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/interrupt.h>
28 #include <linux/kernel_stat.h>
29 #include <linux/delay.h>
30 #include <linux/init.h>
31 #include <linux/spinlock.h>
32 #include <linux/errno.h>
33 #include <linux/hardirq.h>
34 #include <linux/cpu.h>
35 #include <linux/compiler.h>
36 
37 #include <asm/ptrace.h>
38 #include <asm/atomic.h>
39 #include <asm/irq.h>
40 #include <asm/page.h>
41 #include <asm/pgtable.h>
42 #include <asm/sections.h>
43 #include <asm/io.h>
44 #include <asm/prom.h>
45 #include <asm/smp.h>
46 #include <asm/machdep.h>
47 #include <asm/pmac_feature.h>
48 #include <asm/time.h>
49 #include <asm/mpic.h>
50 #include <asm/cacheflush.h>
51 #include <asm/keylargo.h>
52 #include <asm/pmac_low_i2c.h>
53 #include <asm/pmac_pfunc.h>
54 
55 #define DEBUG
56 
57 #ifdef DEBUG
58 #define DBG(fmt...) udbg_printf(fmt)
59 #else
60 #define DBG(fmt...)
61 #endif
62 
63 extern void __secondary_start_pmac_0(void);
64 extern int pmac_pfunc_base_install(void);
65 
66 #ifdef CONFIG_PPC32
67 
68 /* Sync flag for HW tb sync */
69 static volatile int sec_tb_reset = 0;
70 
71 /*
72  * Powersurge (old powermac SMP) support.
73  */
74 
75 /* Addresses for powersurge registers */
76 #define HAMMERHEAD_BASE		0xf8000000
77 #define HHEAD_CONFIG		0x90
78 #define HHEAD_SEC_INTR		0xc0
79 
80 /* register for interrupting the primary processor on the powersurge */
81 /* N.B. this is actually the ethernet ROM! */
82 #define PSURGE_PRI_INTR		0xf3019000
83 
84 /* register for storing the start address for the secondary processor */
85 /* N.B. this is the PCI config space address register for the 1st bridge */
86 #define PSURGE_START		0xf2800000
87 
88 /* Daystar/XLR8 4-CPU card */
89 #define PSURGE_QUAD_REG_ADDR	0xf8800000
90 
91 #define PSURGE_QUAD_IRQ_SET	0
92 #define PSURGE_QUAD_IRQ_CLR	1
93 #define PSURGE_QUAD_IRQ_PRIMARY	2
94 #define PSURGE_QUAD_CKSTOP_CTL	3
95 #define PSURGE_QUAD_PRIMARY_ARB	4
96 #define PSURGE_QUAD_BOARD_ID	6
97 #define PSURGE_QUAD_WHICH_CPU	7
98 #define PSURGE_QUAD_CKSTOP_RDBK	8
99 #define PSURGE_QUAD_RESET_CTL	11
100 
101 #define PSURGE_QUAD_OUT(r, v)	(out_8(quad_base + ((r) << 4) + 4, (v)))
102 #define PSURGE_QUAD_IN(r)	(in_8(quad_base + ((r) << 4) + 4) & 0x0f)
103 #define PSURGE_QUAD_BIS(r, v)	(PSURGE_QUAD_OUT((r), PSURGE_QUAD_IN(r) | (v)))
104 #define PSURGE_QUAD_BIC(r, v)	(PSURGE_QUAD_OUT((r), PSURGE_QUAD_IN(r) & ~(v)))
105 
106 /* virtual addresses for the above */
107 static volatile u8 __iomem *hhead_base;
108 static volatile u8 __iomem *quad_base;
109 static volatile u32 __iomem *psurge_pri_intr;
110 static volatile u8 __iomem *psurge_sec_intr;
111 static volatile u32 __iomem *psurge_start;
112 
113 /* values for psurge_type */
114 #define PSURGE_NONE		-1
115 #define PSURGE_DUAL		0
116 #define PSURGE_QUAD_OKEE	1
117 #define PSURGE_QUAD_COTTON	2
118 #define PSURGE_QUAD_ICEGRASS	3
119 
120 /* what sort of powersurge board we have */
121 static int psurge_type = PSURGE_NONE;
122 
123 /*
124  * Set and clear IPIs for powersurge.
125  */
126 static inline void psurge_set_ipi(int cpu)
127 {
128 	if (psurge_type == PSURGE_NONE)
129 		return;
130 	if (cpu == 0)
131 		in_be32(psurge_pri_intr);
132 	else if (psurge_type == PSURGE_DUAL)
133 		out_8(psurge_sec_intr, 0);
134 	else
135 		PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_SET, 1 << cpu);
136 }
137 
138 static inline void psurge_clr_ipi(int cpu)
139 {
140 	if (cpu > 0) {
141 		switch(psurge_type) {
142 		case PSURGE_DUAL:
143 			out_8(psurge_sec_intr, ~0);
144 		case PSURGE_NONE:
145 			break;
146 		default:
147 			PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_CLR, 1 << cpu);
148 		}
149 	}
150 }
151 
152 /*
153  * On powersurge (old SMP powermac architecture) we don't have
154  * separate IPIs for separate messages like openpic does.  Instead
155  * we have a bitmap for each processor, where a 1 bit means that
156  * the corresponding message is pending for that processor.
157  * Ideally each cpu's entry would be in a different cache line.
158  *  -- paulus.
159  */
160 static unsigned long psurge_smp_message[NR_CPUS];
161 
162 void psurge_smp_message_recv(void)
163 {
164 	int cpu = smp_processor_id();
165 	int msg;
166 
167 	/* clear interrupt */
168 	psurge_clr_ipi(cpu);
169 
170 	if (num_online_cpus() < 2)
171 		return;
172 
173 	/* make sure there is a message there */
174 	for (msg = 0; msg < 4; msg++)
175 		if (test_and_clear_bit(msg, &psurge_smp_message[cpu]))
176 			smp_message_recv(msg);
177 }
178 
179 irqreturn_t psurge_primary_intr(int irq, void *d)
180 {
181 	psurge_smp_message_recv();
182 	return IRQ_HANDLED;
183 }
184 
185 static void smp_psurge_message_pass(int target, int msg)
186 {
187 	int i;
188 
189 	if (num_online_cpus() < 2)
190 		return;
191 
192 	for_each_online_cpu(i) {
193 		if (target == MSG_ALL
194 		    || (target == MSG_ALL_BUT_SELF && i != smp_processor_id())
195 		    || target == i) {
196 			set_bit(msg, &psurge_smp_message[i]);
197 			psurge_set_ipi(i);
198 		}
199 	}
200 }
201 
202 /*
203  * Determine a quad card presence. We read the board ID register, we
204  * force the data bus to change to something else, and we read it again.
205  * It it's stable, then the register probably exist (ugh !)
206  */
207 static int __init psurge_quad_probe(void)
208 {
209 	int type;
210 	unsigned int i;
211 
212 	type = PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID);
213 	if (type < PSURGE_QUAD_OKEE || type > PSURGE_QUAD_ICEGRASS
214 	    || type != PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID))
215 		return PSURGE_DUAL;
216 
217 	/* looks OK, try a slightly more rigorous test */
218 	/* bogus is not necessarily cacheline-aligned,
219 	   though I don't suppose that really matters.  -- paulus */
220 	for (i = 0; i < 100; i++) {
221 		volatile u32 bogus[8];
222 		bogus[(0+i)%8] = 0x00000000;
223 		bogus[(1+i)%8] = 0x55555555;
224 		bogus[(2+i)%8] = 0xFFFFFFFF;
225 		bogus[(3+i)%8] = 0xAAAAAAAA;
226 		bogus[(4+i)%8] = 0x33333333;
227 		bogus[(5+i)%8] = 0xCCCCCCCC;
228 		bogus[(6+i)%8] = 0xCCCCCCCC;
229 		bogus[(7+i)%8] = 0x33333333;
230 		wmb();
231 		asm volatile("dcbf 0,%0" : : "r" (bogus) : "memory");
232 		mb();
233 		if (type != PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID))
234 			return PSURGE_DUAL;
235 	}
236 	return type;
237 }
238 
239 static void __init psurge_quad_init(void)
240 {
241 	int procbits;
242 
243 	if (ppc_md.progress) ppc_md.progress("psurge_quad_init", 0x351);
244 	procbits = ~PSURGE_QUAD_IN(PSURGE_QUAD_WHICH_CPU);
245 	if (psurge_type == PSURGE_QUAD_ICEGRASS)
246 		PSURGE_QUAD_BIS(PSURGE_QUAD_RESET_CTL, procbits);
247 	else
248 		PSURGE_QUAD_BIC(PSURGE_QUAD_CKSTOP_CTL, procbits);
249 	mdelay(33);
250 	out_8(psurge_sec_intr, ~0);
251 	PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_CLR, procbits);
252 	PSURGE_QUAD_BIS(PSURGE_QUAD_RESET_CTL, procbits);
253 	if (psurge_type != PSURGE_QUAD_ICEGRASS)
254 		PSURGE_QUAD_BIS(PSURGE_QUAD_CKSTOP_CTL, procbits);
255 	PSURGE_QUAD_BIC(PSURGE_QUAD_PRIMARY_ARB, procbits);
256 	mdelay(33);
257 	PSURGE_QUAD_BIC(PSURGE_QUAD_RESET_CTL, procbits);
258 	mdelay(33);
259 	PSURGE_QUAD_BIS(PSURGE_QUAD_PRIMARY_ARB, procbits);
260 	mdelay(33);
261 }
262 
263 static int __init smp_psurge_probe(void)
264 {
265 	int i, ncpus;
266 	struct device_node *dn;
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 	dn = of_find_node_by_name(NULL, "hammerhead");
283 	if (dn == NULL)
284 		return 1;
285 	of_node_put(dn);
286 
287 	hhead_base = ioremap(HAMMERHEAD_BASE, 0x800);
288 	quad_base = ioremap(PSURGE_QUAD_REG_ADDR, 1024);
289 	psurge_sec_intr = hhead_base + HHEAD_SEC_INTR;
290 
291 	psurge_type = psurge_quad_probe();
292 	if (psurge_type != PSURGE_DUAL) {
293 		psurge_quad_init();
294 		/* All released cards using this HW design have 4 CPUs */
295 		ncpus = 4;
296 	} else {
297 		iounmap(quad_base);
298 		if ((in_8(hhead_base + HHEAD_CONFIG) & 0x02) == 0) {
299 			/* not a dual-cpu card */
300 			iounmap(hhead_base);
301 			psurge_type = PSURGE_NONE;
302 			return 1;
303 		}
304 		ncpus = 2;
305 	}
306 
307 	psurge_start = ioremap(PSURGE_START, 4);
308 	psurge_pri_intr = ioremap(PSURGE_PRI_INTR, 4);
309 
310 	/*
311 	 * This is necessary because OF doesn't know about the
312 	 * secondary cpu(s), and thus there aren't nodes in the
313 	 * device tree for them, and smp_setup_cpu_maps hasn't
314 	 * set their bits in cpu_possible_map and cpu_present_map.
315 	 */
316 	if (ncpus > NR_CPUS)
317 		ncpus = NR_CPUS;
318 	for (i = 1; i < ncpus ; ++i) {
319 		cpu_set(i, cpu_present_map);
320 		set_hard_smp_processor_id(i, i);
321 	}
322 
323 	if (ppc_md.progress) ppc_md.progress("smp_psurge_probe - done", 0x352);
324 
325 	return ncpus;
326 }
327 
328 static void __init smp_psurge_kick_cpu(int nr)
329 {
330 	unsigned long start = __pa(__secondary_start_pmac_0) + nr * 8;
331 	unsigned long a;
332 	int i;
333 
334 	/* may need to flush here if secondary bats aren't setup */
335 	for (a = KERNELBASE; a < KERNELBASE + 0x800000; a += 32)
336 		asm volatile("dcbf 0,%0" : : "r" (a) : "memory");
337 	asm volatile("sync");
338 
339 	if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu", 0x353);
340 
341 	out_be32(psurge_start, start);
342 	mb();
343 
344 	psurge_set_ipi(nr);
345 	/*
346 	 * We can't use udelay here because the timebase is now frozen.
347 	 */
348 	for (i = 0; i < 2000; ++i)
349 		barrier();
350 	psurge_clr_ipi(nr);
351 
352 	if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu - done", 0x354);
353 }
354 
355 /*
356  * With the dual-cpu powersurge board, the decrementers and timebases
357  * of both cpus are frozen after the secondary cpu is started up,
358  * until we give the secondary cpu another interrupt.  This routine
359  * uses this to get the timebases synchronized.
360  *  -- paulus.
361  */
362 static void __init psurge_dual_sync_tb(int cpu_nr)
363 {
364 	int t;
365 
366 	set_dec(tb_ticks_per_jiffy);
367 	/* XXX fixme */
368 	set_tb(0, 0);
369 
370 	if (cpu_nr > 0) {
371 		mb();
372 		sec_tb_reset = 1;
373 		return;
374 	}
375 
376 	/* wait for the secondary to have reset its TB before proceeding */
377 	for (t = 10000000; t > 0 && !sec_tb_reset; --t)
378 		;
379 
380 	/* now interrupt the secondary, starting both TBs */
381 	psurge_set_ipi(1);
382 }
383 
384 static struct irqaction psurge_irqaction = {
385 	.handler = psurge_primary_intr,
386 	.flags = IRQF_DISABLED,
387 	.mask = CPU_MASK_NONE,
388 	.name = "primary IPI",
389 };
390 
391 static void __init smp_psurge_setup_cpu(int cpu_nr)
392 {
393 
394 	if (cpu_nr == 0) {
395 		/* If we failed to start the second CPU, we should still
396 		 * send it an IPI to start the timebase & DEC or we might
397 		 * have them stuck.
398 		 */
399 		if (num_online_cpus() < 2) {
400 			if (psurge_type == PSURGE_DUAL)
401 				psurge_set_ipi(1);
402 			return;
403 		}
404 		/* reset the entry point so if we get another intr we won't
405 		 * try to startup again */
406 		out_be32(psurge_start, 0x100);
407 		if (setup_irq(30, &psurge_irqaction))
408 			printk(KERN_ERR "Couldn't get primary IPI interrupt");
409 	}
410 
411 	if (psurge_type == PSURGE_DUAL)
412 		psurge_dual_sync_tb(cpu_nr);
413 }
414 
415 void __init smp_psurge_take_timebase(void)
416 {
417 	/* Dummy implementation */
418 }
419 
420 void __init smp_psurge_give_timebase(void)
421 {
422 	/* Dummy implementation */
423 }
424 
425 /* PowerSurge-style Macs */
426 struct smp_ops_t psurge_smp_ops = {
427 	.message_pass	= smp_psurge_message_pass,
428 	.probe		= smp_psurge_probe,
429 	.kick_cpu	= smp_psurge_kick_cpu,
430 	.setup_cpu	= smp_psurge_setup_cpu,
431 	.give_timebase	= smp_psurge_give_timebase,
432 	.take_timebase	= smp_psurge_take_timebase,
433 };
434 #endif /* CONFIG_PPC32 - actually powersurge support */
435 
436 /*
437  * Core 99 and later support
438  */
439 
440 static void (*pmac_tb_freeze)(int freeze);
441 static u64 timebase;
442 static int tb_req;
443 
444 static void smp_core99_give_timebase(void)
445 {
446 	unsigned long flags;
447 
448 	local_irq_save(flags);
449 
450 	while(!tb_req)
451 		barrier();
452 	tb_req = 0;
453 	(*pmac_tb_freeze)(1);
454 	mb();
455 	timebase = get_tb();
456 	mb();
457 	while (timebase)
458 		barrier();
459 	mb();
460 	(*pmac_tb_freeze)(0);
461 	mb();
462 
463 	local_irq_restore(flags);
464 }
465 
466 
467 static void __devinit smp_core99_take_timebase(void)
468 {
469 	unsigned long flags;
470 
471 	local_irq_save(flags);
472 
473 	tb_req = 1;
474 	mb();
475 	while (!timebase)
476 		barrier();
477 	mb();
478 	set_tb(timebase >> 32, timebase & 0xffffffff);
479 	timebase = 0;
480 	mb();
481 	set_dec(tb_ticks_per_jiffy/2);
482 
483 	local_irq_restore(flags);
484 }
485 
486 #ifdef CONFIG_PPC64
487 /*
488  * G5s enable/disable the timebase via an i2c-connected clock chip.
489  */
490 static struct pmac_i2c_bus *pmac_tb_clock_chip_host;
491 static u8 pmac_tb_pulsar_addr;
492 
493 static void smp_core99_cypress_tb_freeze(int freeze)
494 {
495 	u8 data;
496 	int rc;
497 
498 	/* Strangely, the device-tree says address is 0xd2, but darwin
499 	 * accesses 0xd0 ...
500 	 */
501 	pmac_i2c_setmode(pmac_tb_clock_chip_host,
502 			 pmac_i2c_mode_combined);
503 	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
504 			   0xd0 | pmac_i2c_read,
505 			   1, 0x81, &data, 1);
506 	if (rc != 0)
507 		goto bail;
508 
509 	data = (data & 0xf3) | (freeze ? 0x00 : 0x0c);
510 
511        	pmac_i2c_setmode(pmac_tb_clock_chip_host, pmac_i2c_mode_stdsub);
512 	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
513 			   0xd0 | pmac_i2c_write,
514 			   1, 0x81, &data, 1);
515 
516  bail:
517 	if (rc != 0) {
518 		printk("Cypress Timebase %s rc: %d\n",
519 		       freeze ? "freeze" : "unfreeze", rc);
520 		panic("Timebase freeze failed !\n");
521 	}
522 }
523 
524 
525 static void smp_core99_pulsar_tb_freeze(int freeze)
526 {
527 	u8 data;
528 	int rc;
529 
530 	pmac_i2c_setmode(pmac_tb_clock_chip_host,
531 			 pmac_i2c_mode_combined);
532 	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
533 			   pmac_tb_pulsar_addr | pmac_i2c_read,
534 			   1, 0x2e, &data, 1);
535 	if (rc != 0)
536 		goto bail;
537 
538 	data = (data & 0x88) | (freeze ? 0x11 : 0x22);
539 
540 	pmac_i2c_setmode(pmac_tb_clock_chip_host, pmac_i2c_mode_stdsub);
541 	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
542 			   pmac_tb_pulsar_addr | pmac_i2c_write,
543 			   1, 0x2e, &data, 1);
544  bail:
545 	if (rc != 0) {
546 		printk(KERN_ERR "Pulsar Timebase %s rc: %d\n",
547 		       freeze ? "freeze" : "unfreeze", rc);
548 		panic("Timebase freeze failed !\n");
549 	}
550 }
551 
552 static void __init smp_core99_setup_i2c_hwsync(int ncpus)
553 {
554 	struct device_node *cc = NULL;
555 	struct device_node *p;
556 	const char *name = NULL;
557 	const u32 *reg;
558 	int ok;
559 
560 	/* Look for the clock chip */
561 	while ((cc = of_find_node_by_name(cc, "i2c-hwclock")) != NULL) {
562 		p = of_get_parent(cc);
563 		ok = p && of_device_is_compatible(p, "uni-n-i2c");
564 		of_node_put(p);
565 		if (!ok)
566 			continue;
567 
568 		pmac_tb_clock_chip_host = pmac_i2c_find_bus(cc);
569 		if (pmac_tb_clock_chip_host == NULL)
570 			continue;
571 		reg = of_get_property(cc, "reg", NULL);
572 		if (reg == NULL)
573 			continue;
574 		switch (*reg) {
575 		case 0xd2:
576 			if (of_device_is_compatible(cc,"pulsar-legacy-slewing")) {
577 				pmac_tb_freeze = smp_core99_pulsar_tb_freeze;
578 				pmac_tb_pulsar_addr = 0xd2;
579 				name = "Pulsar";
580 			} else if (of_device_is_compatible(cc, "cy28508")) {
581 				pmac_tb_freeze = smp_core99_cypress_tb_freeze;
582 				name = "Cypress";
583 			}
584 			break;
585 		case 0xd4:
586 			pmac_tb_freeze = smp_core99_pulsar_tb_freeze;
587 			pmac_tb_pulsar_addr = 0xd4;
588 			name = "Pulsar";
589 			break;
590 		}
591 		if (pmac_tb_freeze != NULL)
592 			break;
593 	}
594 	if (pmac_tb_freeze != NULL) {
595 		/* Open i2c bus for synchronous access */
596 		if (pmac_i2c_open(pmac_tb_clock_chip_host, 1)) {
597 			printk(KERN_ERR "Failed top open i2c bus for clock"
598 			       " sync, fallback to software sync !\n");
599 			goto no_i2c_sync;
600 		}
601 		printk(KERN_INFO "Processor timebase sync using %s i2c clock\n",
602 		       name);
603 		return;
604 	}
605  no_i2c_sync:
606 	pmac_tb_freeze = NULL;
607 	pmac_tb_clock_chip_host = NULL;
608 }
609 
610 
611 
612 /*
613  * Newer G5s uses a platform function
614  */
615 
616 static void smp_core99_pfunc_tb_freeze(int freeze)
617 {
618 	struct device_node *cpus;
619 	struct pmf_args args;
620 
621 	cpus = of_find_node_by_path("/cpus");
622 	BUG_ON(cpus == NULL);
623 	args.count = 1;
624 	args.u[0].v = !freeze;
625 	pmf_call_function(cpus, "cpu-timebase", &args);
626 	of_node_put(cpus);
627 }
628 
629 #else /* CONFIG_PPC64 */
630 
631 /*
632  * SMP G4 use a GPIO to enable/disable the timebase.
633  */
634 
635 static unsigned int core99_tb_gpio;	/* Timebase freeze GPIO */
636 
637 static void smp_core99_gpio_tb_freeze(int freeze)
638 {
639 	if (freeze)
640 		pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 4);
641 	else
642 		pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 0);
643 	pmac_call_feature(PMAC_FTR_READ_GPIO, NULL, core99_tb_gpio, 0);
644 }
645 
646 
647 #endif /* !CONFIG_PPC64 */
648 
649 /* L2 and L3 cache settings to pass from CPU0 to CPU1 on G4 cpus */
650 volatile static long int core99_l2_cache;
651 volatile static long int core99_l3_cache;
652 
653 static void __devinit core99_init_caches(int cpu)
654 {
655 #ifndef CONFIG_PPC64
656 	if (!cpu_has_feature(CPU_FTR_L2CR))
657 		return;
658 
659 	if (cpu == 0) {
660 		core99_l2_cache = _get_L2CR();
661 		printk("CPU0: L2CR is %lx\n", core99_l2_cache);
662 	} else {
663 		printk("CPU%d: L2CR was %lx\n", cpu, _get_L2CR());
664 		_set_L2CR(0);
665 		_set_L2CR(core99_l2_cache);
666 		printk("CPU%d: L2CR set to %lx\n", cpu, core99_l2_cache);
667 	}
668 
669 	if (!cpu_has_feature(CPU_FTR_L3CR))
670 		return;
671 
672 	if (cpu == 0){
673 		core99_l3_cache = _get_L3CR();
674 		printk("CPU0: L3CR is %lx\n", core99_l3_cache);
675 	} else {
676 		printk("CPU%d: L3CR was %lx\n", cpu, _get_L3CR());
677 		_set_L3CR(0);
678 		_set_L3CR(core99_l3_cache);
679 		printk("CPU%d: L3CR set to %lx\n", cpu, core99_l3_cache);
680 	}
681 #endif /* !CONFIG_PPC64 */
682 }
683 
684 static void __init smp_core99_setup(int ncpus)
685 {
686 #ifdef CONFIG_PPC64
687 
688 	/* i2c based HW sync on some G5s */
689 	if (machine_is_compatible("PowerMac7,2") ||
690 	    machine_is_compatible("PowerMac7,3") ||
691 	    machine_is_compatible("RackMac3,1"))
692 		smp_core99_setup_i2c_hwsync(ncpus);
693 
694 	/* pfunc based HW sync on recent G5s */
695 	if (pmac_tb_freeze == NULL) {
696 		struct device_node *cpus =
697 			of_find_node_by_path("/cpus");
698 		if (cpus &&
699 		    of_get_property(cpus, "platform-cpu-timebase", NULL)) {
700 			pmac_tb_freeze = smp_core99_pfunc_tb_freeze;
701 			printk(KERN_INFO "Processor timebase sync using"
702 			       " platform function\n");
703 		}
704 	}
705 
706 #else /* CONFIG_PPC64 */
707 
708 	/* GPIO based HW sync on ppc32 Core99 */
709 	if (pmac_tb_freeze == NULL && !machine_is_compatible("MacRISC4")) {
710 		struct device_node *cpu;
711 		const u32 *tbprop = NULL;
712 
713 		core99_tb_gpio = KL_GPIO_TB_ENABLE;	/* default value */
714 		cpu = of_find_node_by_type(NULL, "cpu");
715 		if (cpu != NULL) {
716 			tbprop = of_get_property(cpu, "timebase-enable", NULL);
717 			if (tbprop)
718 				core99_tb_gpio = *tbprop;
719 			of_node_put(cpu);
720 		}
721 		pmac_tb_freeze = smp_core99_gpio_tb_freeze;
722 		printk(KERN_INFO "Processor timebase sync using"
723 		       " GPIO 0x%02x\n", core99_tb_gpio);
724 	}
725 
726 #endif /* CONFIG_PPC64 */
727 
728 	/* No timebase sync, fallback to software */
729 	if (pmac_tb_freeze == NULL) {
730 		smp_ops->give_timebase = smp_generic_give_timebase;
731 		smp_ops->take_timebase = smp_generic_take_timebase;
732 		printk(KERN_INFO "Processor timebase sync using software\n");
733 	}
734 
735 #ifndef CONFIG_PPC64
736 	{
737 		int i;
738 
739 		/* XXX should get this from reg properties */
740 		for (i = 1; i < ncpus; ++i)
741 			smp_hw_index[i] = i;
742 	}
743 #endif
744 
745 	/* 32 bits SMP can't NAP */
746 	if (!machine_is_compatible("MacRISC4"))
747 		powersave_nap = 0;
748 }
749 
750 static int __init smp_core99_probe(void)
751 {
752 	struct device_node *cpus;
753 	int ncpus = 0;
754 
755 	if (ppc_md.progress) ppc_md.progress("smp_core99_probe", 0x345);
756 
757 	/* Count CPUs in the device-tree */
758        	for (cpus = NULL; (cpus = of_find_node_by_type(cpus, "cpu")) != NULL;)
759 	       	++ncpus;
760 
761 	printk(KERN_INFO "PowerMac SMP probe found %d cpus\n", ncpus);
762 
763 	/* Nothing more to do if less than 2 of them */
764 	if (ncpus <= 1)
765 		return 1;
766 
767 	/* We need to perform some early initialisations before we can start
768 	 * setting up SMP as we are running before initcalls
769 	 */
770 	pmac_pfunc_base_install();
771 	pmac_i2c_init();
772 
773 	/* Setup various bits like timebase sync method, ability to nap, ... */
774 	smp_core99_setup(ncpus);
775 
776 	/* Install IPIs */
777 	mpic_request_ipis();
778 
779 	/* Collect l2cr and l3cr values from CPU 0 */
780 	core99_init_caches(0);
781 
782 	return ncpus;
783 }
784 
785 static void __devinit smp_core99_kick_cpu(int nr)
786 {
787 	unsigned int save_vector;
788 	unsigned long target, flags;
789 	volatile unsigned int *vector
790 		 = ((volatile unsigned int *)(KERNELBASE+0x100));
791 
792 	if (nr < 0 || nr > 3)
793 		return;
794 
795 	if (ppc_md.progress)
796 		ppc_md.progress("smp_core99_kick_cpu", 0x346);
797 
798 	local_irq_save(flags);
799 
800 	/* Save reset vector */
801 	save_vector = *vector;
802 
803 	/* Setup fake reset vector that does
804 	 *   b __secondary_start_pmac_0 + nr*8 - KERNELBASE
805 	 */
806 	target = (unsigned long) __secondary_start_pmac_0 + nr * 8;
807 	create_branch((unsigned long)vector, target, BRANCH_SET_LINK);
808 
809 	/* Put some life in our friend */
810 	pmac_call_feature(PMAC_FTR_RESET_CPU, NULL, nr, 0);
811 
812 	/* FIXME: We wait a bit for the CPU to take the exception, I should
813 	 * instead wait for the entry code to set something for me. Well,
814 	 * ideally, all that crap will be done in prom.c and the CPU left
815 	 * in a RAM-based wait loop like CHRP.
816 	 */
817 	mdelay(1);
818 
819 	/* Restore our exception vector */
820 	*vector = save_vector;
821 	flush_icache_range((unsigned long) vector, (unsigned long) vector + 4);
822 
823 	local_irq_restore(flags);
824 	if (ppc_md.progress) ppc_md.progress("smp_core99_kick_cpu done", 0x347);
825 }
826 
827 static void __devinit smp_core99_setup_cpu(int cpu_nr)
828 {
829 	/* Setup L2/L3 */
830 	if (cpu_nr != 0)
831 		core99_init_caches(cpu_nr);
832 
833 	/* Setup openpic */
834 	mpic_setup_this_cpu();
835 
836 	if (cpu_nr == 0) {
837 #ifdef CONFIG_PPC64
838 		extern void g5_phy_disable_cpu1(void);
839 
840 		/* Close i2c bus if it was used for tb sync */
841 		if (pmac_tb_clock_chip_host) {
842 			pmac_i2c_close(pmac_tb_clock_chip_host);
843 			pmac_tb_clock_chip_host	= NULL;
844 		}
845 
846 		/* If we didn't start the second CPU, we must take
847 		 * it off the bus
848 		 */
849 		if (machine_is_compatible("MacRISC4") &&
850 		    num_online_cpus() < 2)
851 			g5_phy_disable_cpu1();
852 #endif /* CONFIG_PPC64 */
853 
854 		if (ppc_md.progress)
855 			ppc_md.progress("core99_setup_cpu 0 done", 0x349);
856 	}
857 }
858 
859 
860 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PPC32)
861 
862 int smp_core99_cpu_disable(void)
863 {
864 	cpu_clear(smp_processor_id(), cpu_online_map);
865 
866 	/* XXX reset cpu affinity here */
867 	mpic_cpu_set_priority(0xf);
868 	asm volatile("mtdec %0" : : "r" (0x7fffffff));
869 	mb();
870 	udelay(20);
871 	asm volatile("mtdec %0" : : "r" (0x7fffffff));
872 	return 0;
873 }
874 
875 extern void low_cpu_die(void) __attribute__((noreturn)); /* in sleep.S */
876 static int cpu_dead[NR_CPUS];
877 
878 void cpu_die(void)
879 {
880 	local_irq_disable();
881 	cpu_dead[smp_processor_id()] = 1;
882 	mb();
883 	low_cpu_die();
884 }
885 
886 void smp_core99_cpu_die(unsigned int cpu)
887 {
888 	int timeout;
889 
890 	timeout = 1000;
891 	while (!cpu_dead[cpu]) {
892 		if (--timeout == 0) {
893 			printk("CPU %u refused to die!\n", cpu);
894 			break;
895 		}
896 		msleep(1);
897 	}
898 	cpu_dead[cpu] = 0;
899 }
900 
901 #endif /* CONFIG_HOTPLUG_CPU && CONFIG_PP32 */
902 
903 /* Core99 Macs (dual G4s and G5s) */
904 struct smp_ops_t core99_smp_ops = {
905 	.message_pass	= smp_mpic_message_pass,
906 	.probe		= smp_core99_probe,
907 	.kick_cpu	= smp_core99_kick_cpu,
908 	.setup_cpu	= smp_core99_setup_cpu,
909 	.give_timebase	= smp_core99_give_timebase,
910 	.take_timebase	= smp_core99_take_timebase,
911 #if defined(CONFIG_HOTPLUG_CPU)
912 # if defined(CONFIG_PPC32)
913 	.cpu_disable	= smp_core99_cpu_disable,
914 	.cpu_die	= smp_core99_cpu_die,
915 # endif
916 # if defined(CONFIG_PPC64)
917 	.cpu_disable	= generic_cpu_disable,
918 	.cpu_die	= generic_cpu_die,
919 	/* intentionally do *NOT* assign cpu_enable,
920 	 * the generic code will use kick_cpu then! */
921 # endif
922 #endif
923 };
924