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/code-patching.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 	struct device_node *dn;
268 
269 	/* We don't do SMP on the PPC601 -- paulus */
270 	if (PVR_VER(mfspr(SPRN_PVR)) == 1)
271 		return 1;
272 
273 	/*
274 	 * The powersurge cpu board can be used in the generation
275 	 * of powermacs that have a socket for an upgradeable cpu card,
276 	 * including the 7500, 8500, 9500, 9600.
277 	 * The device tree doesn't tell you if you have 2 cpus because
278 	 * OF doesn't know anything about the 2nd processor.
279 	 * Instead we look for magic bits in magic registers,
280 	 * in the hammerhead memory controller in the case of the
281 	 * dual-cpu powersurge board.  -- paulus.
282 	 */
283 	dn = of_find_node_by_name(NULL, "hammerhead");
284 	if (dn == NULL)
285 		return 1;
286 	of_node_put(dn);
287 
288 	hhead_base = ioremap(HAMMERHEAD_BASE, 0x800);
289 	quad_base = ioremap(PSURGE_QUAD_REG_ADDR, 1024);
290 	psurge_sec_intr = hhead_base + HHEAD_SEC_INTR;
291 
292 	psurge_type = psurge_quad_probe();
293 	if (psurge_type != PSURGE_DUAL) {
294 		psurge_quad_init();
295 		/* All released cards using this HW design have 4 CPUs */
296 		ncpus = 4;
297 	} else {
298 		iounmap(quad_base);
299 		if ((in_8(hhead_base + HHEAD_CONFIG) & 0x02) == 0) {
300 			/* not a dual-cpu card */
301 			iounmap(hhead_base);
302 			psurge_type = PSURGE_NONE;
303 			return 1;
304 		}
305 		ncpus = 2;
306 	}
307 
308 	psurge_start = ioremap(PSURGE_START, 4);
309 	psurge_pri_intr = ioremap(PSURGE_PRI_INTR, 4);
310 
311 	/*
312 	 * This is necessary because OF doesn't know about the
313 	 * secondary cpu(s), and thus there aren't nodes in the
314 	 * device tree for them, and smp_setup_cpu_maps hasn't
315 	 * set their bits in cpu_possible_map and cpu_present_map.
316 	 */
317 	if (ncpus > NR_CPUS)
318 		ncpus = NR_CPUS;
319 	for (i = 1; i < ncpus ; ++i) {
320 		cpu_set(i, cpu_present_map);
321 		set_hard_smp_processor_id(i, i);
322 	}
323 
324 	if (ppc_md.progress) ppc_md.progress("smp_psurge_probe - done", 0x352);
325 
326 	return ncpus;
327 }
328 
329 static void __init smp_psurge_kick_cpu(int nr)
330 {
331 	unsigned long start = __pa(__secondary_start_pmac_0) + nr * 8;
332 	unsigned long a;
333 	int i;
334 
335 	/* may need to flush here if secondary bats aren't setup */
336 	for (a = KERNELBASE; a < KERNELBASE + 0x800000; a += 32)
337 		asm volatile("dcbf 0,%0" : : "r" (a) : "memory");
338 	asm volatile("sync");
339 
340 	if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu", 0x353);
341 
342 	out_be32(psurge_start, start);
343 	mb();
344 
345 	psurge_set_ipi(nr);
346 	/*
347 	 * We can't use udelay here because the timebase is now frozen.
348 	 */
349 	for (i = 0; i < 2000; ++i)
350 		barrier();
351 	psurge_clr_ipi(nr);
352 
353 	if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu - done", 0x354);
354 }
355 
356 /*
357  * With the dual-cpu powersurge board, the decrementers and timebases
358  * of both cpus are frozen after the secondary cpu is started up,
359  * until we give the secondary cpu another interrupt.  This routine
360  * uses this to get the timebases synchronized.
361  *  -- paulus.
362  */
363 static void __init psurge_dual_sync_tb(int cpu_nr)
364 {
365 	int t;
366 
367 	set_dec(tb_ticks_per_jiffy);
368 	/* XXX fixme */
369 	set_tb(0, 0);
370 
371 	if (cpu_nr > 0) {
372 		mb();
373 		sec_tb_reset = 1;
374 		return;
375 	}
376 
377 	/* wait for the secondary to have reset its TB before proceeding */
378 	for (t = 10000000; t > 0 && !sec_tb_reset; --t)
379 		;
380 
381 	/* now interrupt the secondary, starting both TBs */
382 	psurge_set_ipi(1);
383 }
384 
385 static struct irqaction psurge_irqaction = {
386 	.handler = psurge_primary_intr,
387 	.flags = IRQF_DISABLED,
388 	.mask = CPU_MASK_NONE,
389 	.name = "primary IPI",
390 };
391 
392 static void __init smp_psurge_setup_cpu(int cpu_nr)
393 {
394 
395 	if (cpu_nr == 0) {
396 		/* If we failed to start the second CPU, we should still
397 		 * send it an IPI to start the timebase & DEC or we might
398 		 * have them stuck.
399 		 */
400 		if (num_online_cpus() < 2) {
401 			if (psurge_type == PSURGE_DUAL)
402 				psurge_set_ipi(1);
403 			return;
404 		}
405 		/* reset the entry point so if we get another intr we won't
406 		 * try to startup again */
407 		out_be32(psurge_start, 0x100);
408 		if (setup_irq(30, &psurge_irqaction))
409 			printk(KERN_ERR "Couldn't get primary IPI interrupt");
410 	}
411 
412 	if (psurge_type == PSURGE_DUAL)
413 		psurge_dual_sync_tb(cpu_nr);
414 }
415 
416 void __init smp_psurge_take_timebase(void)
417 {
418 	/* Dummy implementation */
419 }
420 
421 void __init smp_psurge_give_timebase(void)
422 {
423 	/* Dummy implementation */
424 }
425 
426 /* PowerSurge-style Macs */
427 struct smp_ops_t psurge_smp_ops = {
428 	.message_pass	= smp_psurge_message_pass,
429 	.probe		= smp_psurge_probe,
430 	.kick_cpu	= smp_psurge_kick_cpu,
431 	.setup_cpu	= smp_psurge_setup_cpu,
432 	.give_timebase	= smp_psurge_give_timebase,
433 	.take_timebase	= smp_psurge_take_timebase,
434 };
435 #endif /* CONFIG_PPC32 - actually powersurge support */
436 
437 /*
438  * Core 99 and later support
439  */
440 
441 static void (*pmac_tb_freeze)(int freeze);
442 static u64 timebase;
443 static int tb_req;
444 
445 static void smp_core99_give_timebase(void)
446 {
447 	unsigned long flags;
448 
449 	local_irq_save(flags);
450 
451 	while(!tb_req)
452 		barrier();
453 	tb_req = 0;
454 	(*pmac_tb_freeze)(1);
455 	mb();
456 	timebase = get_tb();
457 	mb();
458 	while (timebase)
459 		barrier();
460 	mb();
461 	(*pmac_tb_freeze)(0);
462 	mb();
463 
464 	local_irq_restore(flags);
465 }
466 
467 
468 static void __devinit smp_core99_take_timebase(void)
469 {
470 	unsigned long flags;
471 
472 	local_irq_save(flags);
473 
474 	tb_req = 1;
475 	mb();
476 	while (!timebase)
477 		barrier();
478 	mb();
479 	set_tb(timebase >> 32, timebase & 0xffffffff);
480 	timebase = 0;
481 	mb();
482 	set_dec(tb_ticks_per_jiffy/2);
483 
484 	local_irq_restore(flags);
485 }
486 
487 #ifdef CONFIG_PPC64
488 /*
489  * G5s enable/disable the timebase via an i2c-connected clock chip.
490  */
491 static struct pmac_i2c_bus *pmac_tb_clock_chip_host;
492 static u8 pmac_tb_pulsar_addr;
493 
494 static void smp_core99_cypress_tb_freeze(int freeze)
495 {
496 	u8 data;
497 	int rc;
498 
499 	/* Strangely, the device-tree says address is 0xd2, but darwin
500 	 * accesses 0xd0 ...
501 	 */
502 	pmac_i2c_setmode(pmac_tb_clock_chip_host,
503 			 pmac_i2c_mode_combined);
504 	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
505 			   0xd0 | pmac_i2c_read,
506 			   1, 0x81, &data, 1);
507 	if (rc != 0)
508 		goto bail;
509 
510 	data = (data & 0xf3) | (freeze ? 0x00 : 0x0c);
511 
512        	pmac_i2c_setmode(pmac_tb_clock_chip_host, pmac_i2c_mode_stdsub);
513 	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
514 			   0xd0 | pmac_i2c_write,
515 			   1, 0x81, &data, 1);
516 
517  bail:
518 	if (rc != 0) {
519 		printk("Cypress Timebase %s rc: %d\n",
520 		       freeze ? "freeze" : "unfreeze", rc);
521 		panic("Timebase freeze failed !\n");
522 	}
523 }
524 
525 
526 static void smp_core99_pulsar_tb_freeze(int freeze)
527 {
528 	u8 data;
529 	int rc;
530 
531 	pmac_i2c_setmode(pmac_tb_clock_chip_host,
532 			 pmac_i2c_mode_combined);
533 	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
534 			   pmac_tb_pulsar_addr | pmac_i2c_read,
535 			   1, 0x2e, &data, 1);
536 	if (rc != 0)
537 		goto bail;
538 
539 	data = (data & 0x88) | (freeze ? 0x11 : 0x22);
540 
541 	pmac_i2c_setmode(pmac_tb_clock_chip_host, pmac_i2c_mode_stdsub);
542 	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
543 			   pmac_tb_pulsar_addr | pmac_i2c_write,
544 			   1, 0x2e, &data, 1);
545  bail:
546 	if (rc != 0) {
547 		printk(KERN_ERR "Pulsar Timebase %s rc: %d\n",
548 		       freeze ? "freeze" : "unfreeze", rc);
549 		panic("Timebase freeze failed !\n");
550 	}
551 }
552 
553 static void __init smp_core99_setup_i2c_hwsync(int ncpus)
554 {
555 	struct device_node *cc = NULL;
556 	struct device_node *p;
557 	const char *name = NULL;
558 	const u32 *reg;
559 	int ok;
560 
561 	/* Look for the clock chip */
562 	while ((cc = of_find_node_by_name(cc, "i2c-hwclock")) != NULL) {
563 		p = of_get_parent(cc);
564 		ok = p && of_device_is_compatible(p, "uni-n-i2c");
565 		of_node_put(p);
566 		if (!ok)
567 			continue;
568 
569 		pmac_tb_clock_chip_host = pmac_i2c_find_bus(cc);
570 		if (pmac_tb_clock_chip_host == NULL)
571 			continue;
572 		reg = of_get_property(cc, "reg", NULL);
573 		if (reg == NULL)
574 			continue;
575 		switch (*reg) {
576 		case 0xd2:
577 			if (of_device_is_compatible(cc,"pulsar-legacy-slewing")) {
578 				pmac_tb_freeze = smp_core99_pulsar_tb_freeze;
579 				pmac_tb_pulsar_addr = 0xd2;
580 				name = "Pulsar";
581 			} else if (of_device_is_compatible(cc, "cy28508")) {
582 				pmac_tb_freeze = smp_core99_cypress_tb_freeze;
583 				name = "Cypress";
584 			}
585 			break;
586 		case 0xd4:
587 			pmac_tb_freeze = smp_core99_pulsar_tb_freeze;
588 			pmac_tb_pulsar_addr = 0xd4;
589 			name = "Pulsar";
590 			break;
591 		}
592 		if (pmac_tb_freeze != NULL)
593 			break;
594 	}
595 	if (pmac_tb_freeze != NULL) {
596 		/* Open i2c bus for synchronous access */
597 		if (pmac_i2c_open(pmac_tb_clock_chip_host, 1)) {
598 			printk(KERN_ERR "Failed top open i2c bus for clock"
599 			       " sync, fallback to software sync !\n");
600 			goto no_i2c_sync;
601 		}
602 		printk(KERN_INFO "Processor timebase sync using %s i2c clock\n",
603 		       name);
604 		return;
605 	}
606  no_i2c_sync:
607 	pmac_tb_freeze = NULL;
608 	pmac_tb_clock_chip_host = NULL;
609 }
610 
611 
612 
613 /*
614  * Newer G5s uses a platform function
615  */
616 
617 static void smp_core99_pfunc_tb_freeze(int freeze)
618 {
619 	struct device_node *cpus;
620 	struct pmf_args args;
621 
622 	cpus = of_find_node_by_path("/cpus");
623 	BUG_ON(cpus == NULL);
624 	args.count = 1;
625 	args.u[0].v = !freeze;
626 	pmf_call_function(cpus, "cpu-timebase", &args);
627 	of_node_put(cpus);
628 }
629 
630 #else /* CONFIG_PPC64 */
631 
632 /*
633  * SMP G4 use a GPIO to enable/disable the timebase.
634  */
635 
636 static unsigned int core99_tb_gpio;	/* Timebase freeze GPIO */
637 
638 static void smp_core99_gpio_tb_freeze(int freeze)
639 {
640 	if (freeze)
641 		pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 4);
642 	else
643 		pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 0);
644 	pmac_call_feature(PMAC_FTR_READ_GPIO, NULL, core99_tb_gpio, 0);
645 }
646 
647 
648 #endif /* !CONFIG_PPC64 */
649 
650 /* L2 and L3 cache settings to pass from CPU0 to CPU1 on G4 cpus */
651 volatile static long int core99_l2_cache;
652 volatile static long int core99_l3_cache;
653 
654 static void __devinit core99_init_caches(int cpu)
655 {
656 #ifndef CONFIG_PPC64
657 	if (!cpu_has_feature(CPU_FTR_L2CR))
658 		return;
659 
660 	if (cpu == 0) {
661 		core99_l2_cache = _get_L2CR();
662 		printk("CPU0: L2CR is %lx\n", core99_l2_cache);
663 	} else {
664 		printk("CPU%d: L2CR was %lx\n", cpu, _get_L2CR());
665 		_set_L2CR(0);
666 		_set_L2CR(core99_l2_cache);
667 		printk("CPU%d: L2CR set to %lx\n", cpu, core99_l2_cache);
668 	}
669 
670 	if (!cpu_has_feature(CPU_FTR_L3CR))
671 		return;
672 
673 	if (cpu == 0){
674 		core99_l3_cache = _get_L3CR();
675 		printk("CPU0: L3CR is %lx\n", core99_l3_cache);
676 	} else {
677 		printk("CPU%d: L3CR was %lx\n", cpu, _get_L3CR());
678 		_set_L3CR(0);
679 		_set_L3CR(core99_l3_cache);
680 		printk("CPU%d: L3CR set to %lx\n", cpu, core99_l3_cache);
681 	}
682 #endif /* !CONFIG_PPC64 */
683 }
684 
685 static void __init smp_core99_setup(int ncpus)
686 {
687 #ifdef CONFIG_PPC64
688 
689 	/* i2c based HW sync on some G5s */
690 	if (machine_is_compatible("PowerMac7,2") ||
691 	    machine_is_compatible("PowerMac7,3") ||
692 	    machine_is_compatible("RackMac3,1"))
693 		smp_core99_setup_i2c_hwsync(ncpus);
694 
695 	/* pfunc based HW sync on recent G5s */
696 	if (pmac_tb_freeze == NULL) {
697 		struct device_node *cpus =
698 			of_find_node_by_path("/cpus");
699 		if (cpus &&
700 		    of_get_property(cpus, "platform-cpu-timebase", NULL)) {
701 			pmac_tb_freeze = smp_core99_pfunc_tb_freeze;
702 			printk(KERN_INFO "Processor timebase sync using"
703 			       " platform function\n");
704 		}
705 	}
706 
707 #else /* CONFIG_PPC64 */
708 
709 	/* GPIO based HW sync on ppc32 Core99 */
710 	if (pmac_tb_freeze == NULL && !machine_is_compatible("MacRISC4")) {
711 		struct device_node *cpu;
712 		const u32 *tbprop = NULL;
713 
714 		core99_tb_gpio = KL_GPIO_TB_ENABLE;	/* default value */
715 		cpu = of_find_node_by_type(NULL, "cpu");
716 		if (cpu != NULL) {
717 			tbprop = of_get_property(cpu, "timebase-enable", NULL);
718 			if (tbprop)
719 				core99_tb_gpio = *tbprop;
720 			of_node_put(cpu);
721 		}
722 		pmac_tb_freeze = smp_core99_gpio_tb_freeze;
723 		printk(KERN_INFO "Processor timebase sync using"
724 		       " GPIO 0x%02x\n", core99_tb_gpio);
725 	}
726 
727 #endif /* CONFIG_PPC64 */
728 
729 	/* No timebase sync, fallback to software */
730 	if (pmac_tb_freeze == NULL) {
731 		smp_ops->give_timebase = smp_generic_give_timebase;
732 		smp_ops->take_timebase = smp_generic_take_timebase;
733 		printk(KERN_INFO "Processor timebase sync using software\n");
734 	}
735 
736 #ifndef CONFIG_PPC64
737 	{
738 		int i;
739 
740 		/* XXX should get this from reg properties */
741 		for (i = 1; i < ncpus; ++i)
742 			smp_hw_index[i] = i;
743 	}
744 #endif
745 
746 	/* 32 bits SMP can't NAP */
747 	if (!machine_is_compatible("MacRISC4"))
748 		powersave_nap = 0;
749 }
750 
751 static int __init smp_core99_probe(void)
752 {
753 	struct device_node *cpus;
754 	int ncpus = 0;
755 
756 	if (ppc_md.progress) ppc_md.progress("smp_core99_probe", 0x345);
757 
758 	/* Count CPUs in the device-tree */
759        	for (cpus = NULL; (cpus = of_find_node_by_type(cpus, "cpu")) != NULL;)
760 	       	++ncpus;
761 
762 	printk(KERN_INFO "PowerMac SMP probe found %d cpus\n", ncpus);
763 
764 	/* Nothing more to do if less than 2 of them */
765 	if (ncpus <= 1)
766 		return 1;
767 
768 	/* We need to perform some early initialisations before we can start
769 	 * setting up SMP as we are running before initcalls
770 	 */
771 	pmac_pfunc_base_install();
772 	pmac_i2c_init();
773 
774 	/* Setup various bits like timebase sync method, ability to nap, ... */
775 	smp_core99_setup(ncpus);
776 
777 	/* Install IPIs */
778 	mpic_request_ipis();
779 
780 	/* Collect l2cr and l3cr values from CPU 0 */
781 	core99_init_caches(0);
782 
783 	return ncpus;
784 }
785 
786 static void __devinit smp_core99_kick_cpu(int nr)
787 {
788 	unsigned int save_vector;
789 	unsigned long target, flags;
790 	unsigned int *vector = (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 	patch_branch(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