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/config.h>
25 #include <linux/kernel.h>
26 #include <linux/sched.h>
27 #include <linux/smp.h>
28 #include <linux/smp_lock.h>
29 #include <linux/interrupt.h>
30 #include <linux/kernel_stat.h>
31 #include <linux/delay.h>
32 #include <linux/init.h>
33 #include <linux/spinlock.h>
34 #include <linux/errno.h>
35 #include <linux/hardirq.h>
36 #include <linux/cpu.h>
37 #include <linux/compiler.h>
38 
39 #include <asm/ptrace.h>
40 #include <asm/atomic.h>
41 #include <asm/irq.h>
42 #include <asm/page.h>
43 #include <asm/pgtable.h>
44 #include <asm/sections.h>
45 #include <asm/io.h>
46 #include <asm/prom.h>
47 #include <asm/smp.h>
48 #include <asm/machdep.h>
49 #include <asm/pmac_feature.h>
50 #include <asm/time.h>
51 #include <asm/mpic.h>
52 #include <asm/cacheflush.h>
53 #include <asm/keylargo.h>
54 #include <asm/pmac_low_i2c.h>
55 #include <asm/pmac_pfunc.h>
56 
57 #define DEBUG
58 
59 #ifdef DEBUG
60 #define DBG(fmt...) udbg_printf(fmt)
61 #else
62 #define DBG(fmt...)
63 #endif
64 
65 extern void __secondary_start_pmac_0(void);
66 extern int pmac_pfunc_base_install(void);
67 
68 #ifdef CONFIG_PPC32
69 
70 /* Sync flag for HW tb sync */
71 static volatile int sec_tb_reset = 0;
72 
73 /*
74  * Powersurge (old powermac SMP) support.
75  */
76 
77 /* Addresses for powersurge registers */
78 #define HAMMERHEAD_BASE		0xf8000000
79 #define HHEAD_CONFIG		0x90
80 #define HHEAD_SEC_INTR		0xc0
81 
82 /* register for interrupting the primary processor on the powersurge */
83 /* N.B. this is actually the ethernet ROM! */
84 #define PSURGE_PRI_INTR		0xf3019000
85 
86 /* register for storing the start address for the secondary processor */
87 /* N.B. this is the PCI config space address register for the 1st bridge */
88 #define PSURGE_START		0xf2800000
89 
90 /* Daystar/XLR8 4-CPU card */
91 #define PSURGE_QUAD_REG_ADDR	0xf8800000
92 
93 #define PSURGE_QUAD_IRQ_SET	0
94 #define PSURGE_QUAD_IRQ_CLR	1
95 #define PSURGE_QUAD_IRQ_PRIMARY	2
96 #define PSURGE_QUAD_CKSTOP_CTL	3
97 #define PSURGE_QUAD_PRIMARY_ARB	4
98 #define PSURGE_QUAD_BOARD_ID	6
99 #define PSURGE_QUAD_WHICH_CPU	7
100 #define PSURGE_QUAD_CKSTOP_RDBK	8
101 #define PSURGE_QUAD_RESET_CTL	11
102 
103 #define PSURGE_QUAD_OUT(r, v)	(out_8(quad_base + ((r) << 4) + 4, (v)))
104 #define PSURGE_QUAD_IN(r)	(in_8(quad_base + ((r) << 4) + 4) & 0x0f)
105 #define PSURGE_QUAD_BIS(r, v)	(PSURGE_QUAD_OUT((r), PSURGE_QUAD_IN(r) | (v)))
106 #define PSURGE_QUAD_BIC(r, v)	(PSURGE_QUAD_OUT((r), PSURGE_QUAD_IN(r) & ~(v)))
107 
108 /* virtual addresses for the above */
109 static volatile u8 __iomem *hhead_base;
110 static volatile u8 __iomem *quad_base;
111 static volatile u32 __iomem *psurge_pri_intr;
112 static volatile u8 __iomem *psurge_sec_intr;
113 static volatile u32 __iomem *psurge_start;
114 
115 /* values for psurge_type */
116 #define PSURGE_NONE		-1
117 #define PSURGE_DUAL		0
118 #define PSURGE_QUAD_OKEE	1
119 #define PSURGE_QUAD_COTTON	2
120 #define PSURGE_QUAD_ICEGRASS	3
121 
122 /* what sort of powersurge board we have */
123 static int psurge_type = PSURGE_NONE;
124 
125 /*
126  * Set and clear IPIs for powersurge.
127  */
128 static inline void psurge_set_ipi(int cpu)
129 {
130 	if (psurge_type == PSURGE_NONE)
131 		return;
132 	if (cpu == 0)
133 		in_be32(psurge_pri_intr);
134 	else if (psurge_type == PSURGE_DUAL)
135 		out_8(psurge_sec_intr, 0);
136 	else
137 		PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_SET, 1 << cpu);
138 }
139 
140 static inline void psurge_clr_ipi(int cpu)
141 {
142 	if (cpu > 0) {
143 		switch(psurge_type) {
144 		case PSURGE_DUAL:
145 			out_8(psurge_sec_intr, ~0);
146 		case PSURGE_NONE:
147 			break;
148 		default:
149 			PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_CLR, 1 << cpu);
150 		}
151 	}
152 }
153 
154 /*
155  * On powersurge (old SMP powermac architecture) we don't have
156  * separate IPIs for separate messages like openpic does.  Instead
157  * we have a bitmap for each processor, where a 1 bit means that
158  * the corresponding message is pending for that processor.
159  * Ideally each cpu's entry would be in a different cache line.
160  *  -- paulus.
161  */
162 static unsigned long psurge_smp_message[NR_CPUS];
163 
164 void psurge_smp_message_recv(struct pt_regs *regs)
165 {
166 	int cpu = smp_processor_id();
167 	int msg;
168 
169 	/* clear interrupt */
170 	psurge_clr_ipi(cpu);
171 
172 	if (num_online_cpus() < 2)
173 		return;
174 
175 	/* make sure there is a message there */
176 	for (msg = 0; msg < 4; msg++)
177 		if (test_and_clear_bit(msg, &psurge_smp_message[cpu]))
178 			smp_message_recv(msg, regs);
179 }
180 
181 irqreturn_t psurge_primary_intr(int irq, void *d, struct pt_regs *regs)
182 {
183 	psurge_smp_message_recv(regs);
184 	return IRQ_HANDLED;
185 }
186 
187 static void smp_psurge_message_pass(int target, int msg)
188 {
189 	int i;
190 
191 	if (num_online_cpus() < 2)
192 		return;
193 
194 	for (i = 0; i < NR_CPUS; i++) {
195 		if (!cpu_online(i))
196 			continue;
197 		if (target == MSG_ALL
198 		    || (target == MSG_ALL_BUT_SELF && i != smp_processor_id())
199 		    || target == i) {
200 			set_bit(msg, &psurge_smp_message[i]);
201 			psurge_set_ipi(i);
202 		}
203 	}
204 }
205 
206 /*
207  * Determine a quad card presence. We read the board ID register, we
208  * force the data bus to change to something else, and we read it again.
209  * It it's stable, then the register probably exist (ugh !)
210  */
211 static int __init psurge_quad_probe(void)
212 {
213 	int type;
214 	unsigned int i;
215 
216 	type = PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID);
217 	if (type < PSURGE_QUAD_OKEE || type > PSURGE_QUAD_ICEGRASS
218 	    || type != PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID))
219 		return PSURGE_DUAL;
220 
221 	/* looks OK, try a slightly more rigorous test */
222 	/* bogus is not necessarily cacheline-aligned,
223 	   though I don't suppose that really matters.  -- paulus */
224 	for (i = 0; i < 100; i++) {
225 		volatile u32 bogus[8];
226 		bogus[(0+i)%8] = 0x00000000;
227 		bogus[(1+i)%8] = 0x55555555;
228 		bogus[(2+i)%8] = 0xFFFFFFFF;
229 		bogus[(3+i)%8] = 0xAAAAAAAA;
230 		bogus[(4+i)%8] = 0x33333333;
231 		bogus[(5+i)%8] = 0xCCCCCCCC;
232 		bogus[(6+i)%8] = 0xCCCCCCCC;
233 		bogus[(7+i)%8] = 0x33333333;
234 		wmb();
235 		asm volatile("dcbf 0,%0" : : "r" (bogus) : "memory");
236 		mb();
237 		if (type != PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID))
238 			return PSURGE_DUAL;
239 	}
240 	return type;
241 }
242 
243 static void __init psurge_quad_init(void)
244 {
245 	int procbits;
246 
247 	if (ppc_md.progress) ppc_md.progress("psurge_quad_init", 0x351);
248 	procbits = ~PSURGE_QUAD_IN(PSURGE_QUAD_WHICH_CPU);
249 	if (psurge_type == PSURGE_QUAD_ICEGRASS)
250 		PSURGE_QUAD_BIS(PSURGE_QUAD_RESET_CTL, procbits);
251 	else
252 		PSURGE_QUAD_BIC(PSURGE_QUAD_CKSTOP_CTL, procbits);
253 	mdelay(33);
254 	out_8(psurge_sec_intr, ~0);
255 	PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_CLR, procbits);
256 	PSURGE_QUAD_BIS(PSURGE_QUAD_RESET_CTL, procbits);
257 	if (psurge_type != PSURGE_QUAD_ICEGRASS)
258 		PSURGE_QUAD_BIS(PSURGE_QUAD_CKSTOP_CTL, procbits);
259 	PSURGE_QUAD_BIC(PSURGE_QUAD_PRIMARY_ARB, procbits);
260 	mdelay(33);
261 	PSURGE_QUAD_BIC(PSURGE_QUAD_RESET_CTL, procbits);
262 	mdelay(33);
263 	PSURGE_QUAD_BIS(PSURGE_QUAD_PRIMARY_ARB, procbits);
264 	mdelay(33);
265 }
266 
267 static int __init smp_psurge_probe(void)
268 {
269 	int i, ncpus;
270 
271 	/* We don't do SMP on the PPC601 -- paulus */
272 	if (PVR_VER(mfspr(SPRN_PVR)) == 1)
273 		return 1;
274 
275 	/*
276 	 * The powersurge cpu board can be used in the generation
277 	 * of powermacs that have a socket for an upgradeable cpu card,
278 	 * including the 7500, 8500, 9500, 9600.
279 	 * The device tree doesn't tell you if you have 2 cpus because
280 	 * OF doesn't know anything about the 2nd processor.
281 	 * Instead we look for magic bits in magic registers,
282 	 * in the hammerhead memory controller in the case of the
283 	 * dual-cpu powersurge board.  -- paulus.
284 	 */
285 	if (find_devices("hammerhead") == NULL)
286 		return 1;
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 		cpu_set(i, cpu_possible_map);
322 		set_hard_smp_processor_id(i, i);
323 	}
324 
325 	if (ppc_md.progress) ppc_md.progress("smp_psurge_probe - done", 0x352);
326 
327 	return ncpus;
328 }
329 
330 static void __init smp_psurge_kick_cpu(int nr)
331 {
332 	unsigned long start = __pa(__secondary_start_pmac_0) + nr * 8;
333 	unsigned long a;
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 	udelay(10);
347 	psurge_clr_ipi(nr);
348 
349 	if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu - done", 0x354);
350 }
351 
352 /*
353  * With the dual-cpu powersurge board, the decrementers and timebases
354  * of both cpus are frozen after the secondary cpu is started up,
355  * until we give the secondary cpu another interrupt.  This routine
356  * uses this to get the timebases synchronized.
357  *  -- paulus.
358  */
359 static void __init psurge_dual_sync_tb(int cpu_nr)
360 {
361 	int t;
362 
363 	set_dec(tb_ticks_per_jiffy);
364 	/* XXX fixme */
365 	set_tb(0, 0);
366 
367 	if (cpu_nr > 0) {
368 		mb();
369 		sec_tb_reset = 1;
370 		return;
371 	}
372 
373 	/* wait for the secondary to have reset its TB before proceeding */
374 	for (t = 10000000; t > 0 && !sec_tb_reset; --t)
375 		;
376 
377 	/* now interrupt the secondary, starting both TBs */
378 	psurge_set_ipi(1);
379 }
380 
381 static struct irqaction psurge_irqaction = {
382 	.handler = psurge_primary_intr,
383 	.flags = SA_INTERRUPT,
384 	.mask = CPU_MASK_NONE,
385 	.name = "primary IPI",
386 };
387 
388 static void __init smp_psurge_setup_cpu(int cpu_nr)
389 {
390 
391 	if (cpu_nr == 0) {
392 		/* If we failed to start the second CPU, we should still
393 		 * send it an IPI to start the timebase & DEC or we might
394 		 * have them stuck.
395 		 */
396 		if (num_online_cpus() < 2) {
397 			if (psurge_type == PSURGE_DUAL)
398 				psurge_set_ipi(1);
399 			return;
400 		}
401 		/* reset the entry point so if we get another intr we won't
402 		 * try to startup again */
403 		out_be32(psurge_start, 0x100);
404 		if (setup_irq(30, &psurge_irqaction))
405 			printk(KERN_ERR "Couldn't get primary IPI interrupt");
406 	}
407 
408 	if (psurge_type == PSURGE_DUAL)
409 		psurge_dual_sync_tb(cpu_nr);
410 }
411 
412 void __init smp_psurge_take_timebase(void)
413 {
414 	/* Dummy implementation */
415 }
416 
417 void __init smp_psurge_give_timebase(void)
418 {
419 	/* Dummy implementation */
420 }
421 
422 /* PowerSurge-style Macs */
423 struct smp_ops_t psurge_smp_ops = {
424 	.message_pass	= smp_psurge_message_pass,
425 	.probe		= smp_psurge_probe,
426 	.kick_cpu	= smp_psurge_kick_cpu,
427 	.setup_cpu	= smp_psurge_setup_cpu,
428 	.give_timebase	= smp_psurge_give_timebase,
429 	.take_timebase	= smp_psurge_take_timebase,
430 };
431 #endif /* CONFIG_PPC32 - actually powersurge support */
432 
433 /*
434  * Core 99 and later support
435  */
436 
437 static void (*pmac_tb_freeze)(int freeze);
438 static unsigned long timebase;
439 static int tb_req;
440 
441 static void smp_core99_give_timebase(void)
442 {
443 	unsigned long flags;
444 
445 	local_irq_save(flags);
446 
447 	while(!tb_req)
448 		barrier();
449 	tb_req = 0;
450 	(*pmac_tb_freeze)(1);
451 	mb();
452 	timebase = get_tb();
453 	mb();
454 	while (timebase)
455 		barrier();
456 	mb();
457 	(*pmac_tb_freeze)(0);
458 	mb();
459 
460 	local_irq_restore(flags);
461 }
462 
463 
464 static void __devinit smp_core99_take_timebase(void)
465 {
466 	unsigned long flags;
467 
468 	local_irq_save(flags);
469 
470 	tb_req = 1;
471 	mb();
472 	while (!timebase)
473 		barrier();
474 	mb();
475 	set_tb(timebase >> 32, timebase & 0xffffffff);
476 	timebase = 0;
477 	mb();
478 	set_dec(tb_ticks_per_jiffy/2);
479 
480 	local_irq_restore(flags);
481 }
482 
483 #ifdef CONFIG_PPC64
484 /*
485  * G5s enable/disable the timebase via an i2c-connected clock chip.
486  */
487 static struct pmac_i2c_bus *pmac_tb_clock_chip_host;
488 static u8 pmac_tb_pulsar_addr;
489 
490 static void smp_core99_cypress_tb_freeze(int freeze)
491 {
492 	u8 data;
493 	int rc;
494 
495 	/* Strangely, the device-tree says address is 0xd2, but darwin
496 	 * accesses 0xd0 ...
497 	 */
498 	pmac_i2c_setmode(pmac_tb_clock_chip_host,
499 			 pmac_i2c_mode_combined);
500 	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
501 			   0xd0 | pmac_i2c_read,
502 			   1, 0x81, &data, 1);
503 	if (rc != 0)
504 		goto bail;
505 
506 	data = (data & 0xf3) | (freeze ? 0x00 : 0x0c);
507 
508        	pmac_i2c_setmode(pmac_tb_clock_chip_host, pmac_i2c_mode_stdsub);
509 	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
510 			   0xd0 | pmac_i2c_write,
511 			   1, 0x81, &data, 1);
512 
513  bail:
514 	if (rc != 0) {
515 		printk("Cypress Timebase %s rc: %d\n",
516 		       freeze ? "freeze" : "unfreeze", rc);
517 		panic("Timebase freeze failed !\n");
518 	}
519 }
520 
521 
522 static void smp_core99_pulsar_tb_freeze(int freeze)
523 {
524 	u8 data;
525 	int rc;
526 
527 	pmac_i2c_setmode(pmac_tb_clock_chip_host,
528 			 pmac_i2c_mode_combined);
529 	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
530 			   pmac_tb_pulsar_addr | pmac_i2c_read,
531 			   1, 0x2e, &data, 1);
532 	if (rc != 0)
533 		goto bail;
534 
535 	data = (data & 0x88) | (freeze ? 0x11 : 0x22);
536 
537 	pmac_i2c_setmode(pmac_tb_clock_chip_host, pmac_i2c_mode_stdsub);
538 	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
539 			   pmac_tb_pulsar_addr | pmac_i2c_write,
540 			   1, 0x2e, &data, 1);
541  bail:
542 	if (rc != 0) {
543 		printk(KERN_ERR "Pulsar Timebase %s rc: %d\n",
544 		       freeze ? "freeze" : "unfreeze", rc);
545 		panic("Timebase freeze failed !\n");
546 	}
547 }
548 
549 static void __init smp_core99_setup_i2c_hwsync(int ncpus)
550 {
551 	struct device_node *cc = NULL;
552 	struct device_node *p;
553 	const char *name = NULL;
554 	u32 *reg;
555 	int ok;
556 
557 	/* Look for the clock chip */
558 	while ((cc = of_find_node_by_name(cc, "i2c-hwclock")) != NULL) {
559 		p = of_get_parent(cc);
560 		ok = p && device_is_compatible(p, "uni-n-i2c");
561 		of_node_put(p);
562 		if (!ok)
563 			continue;
564 
565 		pmac_tb_clock_chip_host = pmac_i2c_find_bus(cc);
566 		if (pmac_tb_clock_chip_host == NULL)
567 			continue;
568 		reg = (u32 *)get_property(cc, "reg", NULL);
569 		if (reg == NULL)
570 			continue;
571 		switch (*reg) {
572 		case 0xd2:
573 			if (device_is_compatible(cc,"pulsar-legacy-slewing")) {
574 				pmac_tb_freeze = smp_core99_pulsar_tb_freeze;
575 				pmac_tb_pulsar_addr = 0xd2;
576 				name = "Pulsar";
577 			} else if (device_is_compatible(cc, "cy28508")) {
578 				pmac_tb_freeze = smp_core99_cypress_tb_freeze;
579 				name = "Cypress";
580 			}
581 			break;
582 		case 0xd4:
583 			pmac_tb_freeze = smp_core99_pulsar_tb_freeze;
584 			pmac_tb_pulsar_addr = 0xd4;
585 			name = "Pulsar";
586 			break;
587 		}
588 		if (pmac_tb_freeze != NULL)
589 			break;
590 	}
591 	if (pmac_tb_freeze != NULL) {
592 		/* Open i2c bus for synchronous access */
593 		if (pmac_i2c_open(pmac_tb_clock_chip_host, 1)) {
594 			printk(KERN_ERR "Failed top open i2c bus for clock"
595 			       " sync, fallback to software sync !\n");
596 			goto no_i2c_sync;
597 		}
598 		printk(KERN_INFO "Processor timebase sync using %s i2c clock\n",
599 		       name);
600 		return;
601 	}
602  no_i2c_sync:
603 	pmac_tb_freeze = NULL;
604 	pmac_tb_clock_chip_host = NULL;
605 }
606 
607 
608 
609 /*
610  * Newer G5s uses a platform function
611  */
612 
613 static void smp_core99_pfunc_tb_freeze(int freeze)
614 {
615 	struct device_node *cpus;
616 	struct pmf_args args;
617 
618 	cpus = of_find_node_by_path("/cpus");
619 	BUG_ON(cpus == NULL);
620 	args.count = 1;
621 	args.u[0].v = !freeze;
622 	pmf_call_function(cpus, "cpu-timebase", &args);
623 	of_node_put(cpus);
624 }
625 
626 #else /* CONFIG_PPC64 */
627 
628 /*
629  * SMP G4 use a GPIO to enable/disable the timebase.
630  */
631 
632 static unsigned int core99_tb_gpio;	/* Timebase freeze GPIO */
633 
634 static void smp_core99_gpio_tb_freeze(int freeze)
635 {
636 	if (freeze)
637 		pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 4);
638 	else
639 		pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 0);
640 	pmac_call_feature(PMAC_FTR_READ_GPIO, NULL, core99_tb_gpio, 0);
641 }
642 
643 
644 #endif /* !CONFIG_PPC64 */
645 
646 /* L2 and L3 cache settings to pass from CPU0 to CPU1 on G4 cpus */
647 volatile static long int core99_l2_cache;
648 volatile static long int core99_l3_cache;
649 
650 static void __devinit core99_init_caches(int cpu)
651 {
652 #ifndef CONFIG_PPC64
653 	if (!cpu_has_feature(CPU_FTR_L2CR))
654 		return;
655 
656 	if (cpu == 0) {
657 		core99_l2_cache = _get_L2CR();
658 		printk("CPU0: L2CR is %lx\n", core99_l2_cache);
659 	} else {
660 		printk("CPU%d: L2CR was %lx\n", cpu, _get_L2CR());
661 		_set_L2CR(0);
662 		_set_L2CR(core99_l2_cache);
663 		printk("CPU%d: L2CR set to %lx\n", cpu, core99_l2_cache);
664 	}
665 
666 	if (!cpu_has_feature(CPU_FTR_L3CR))
667 		return;
668 
669 	if (cpu == 0){
670 		core99_l3_cache = _get_L3CR();
671 		printk("CPU0: L3CR is %lx\n", core99_l3_cache);
672 	} else {
673 		printk("CPU%d: L3CR was %lx\n", cpu, _get_L3CR());
674 		_set_L3CR(0);
675 		_set_L3CR(core99_l3_cache);
676 		printk("CPU%d: L3CR set to %lx\n", cpu, core99_l3_cache);
677 	}
678 #endif /* !CONFIG_PPC64 */
679 }
680 
681 static void __init smp_core99_setup(int ncpus)
682 {
683 #ifdef CONFIG_PPC64
684 
685 	/* i2c based HW sync on some G5s */
686 	if (machine_is_compatible("PowerMac7,2") ||
687 	    machine_is_compatible("PowerMac7,3") ||
688 	    machine_is_compatible("RackMac3,1"))
689 		smp_core99_setup_i2c_hwsync(ncpus);
690 
691 	/* pfunc based HW sync on recent G5s */
692 	if (pmac_tb_freeze == NULL) {
693 		struct device_node *cpus =
694 			of_find_node_by_path("/cpus");
695 		if (cpus &&
696 		    get_property(cpus, "platform-cpu-timebase", NULL)) {
697 			pmac_tb_freeze = smp_core99_pfunc_tb_freeze;
698 			printk(KERN_INFO "Processor timebase sync using"
699 			       " platform function\n");
700 		}
701 	}
702 
703 #else /* CONFIG_PPC64 */
704 
705 	/* GPIO based HW sync on ppc32 Core99 */
706 	if (pmac_tb_freeze == NULL && !machine_is_compatible("MacRISC4")) {
707 		struct device_node *cpu;
708 		u32 *tbprop = NULL;
709 
710 		core99_tb_gpio = KL_GPIO_TB_ENABLE;	/* default value */
711 		cpu = of_find_node_by_type(NULL, "cpu");
712 		if (cpu != NULL) {
713 			tbprop = (u32 *)get_property(cpu, "timebase-enable",
714 						     NULL);
715 			if (tbprop)
716 				core99_tb_gpio = *tbprop;
717 			of_node_put(cpu);
718 		}
719 		pmac_tb_freeze = smp_core99_gpio_tb_freeze;
720 		printk(KERN_INFO "Processor timebase sync using"
721 		       " GPIO 0x%02x\n", core99_tb_gpio);
722 	}
723 
724 #endif /* CONFIG_PPC64 */
725 
726 	/* No timebase sync, fallback to software */
727 	if (pmac_tb_freeze == NULL) {
728 		smp_ops->give_timebase = smp_generic_give_timebase;
729 		smp_ops->take_timebase = smp_generic_take_timebase;
730 		printk(KERN_INFO "Processor timebase sync using software\n");
731 	}
732 
733 #ifndef CONFIG_PPC64
734 	{
735 		int i;
736 
737 		/* XXX should get this from reg properties */
738 		for (i = 1; i < ncpus; ++i)
739 			smp_hw_index[i] = i;
740 	}
741 #endif
742 
743 	/* 32 bits SMP can't NAP */
744 	if (!machine_is_compatible("MacRISC4"))
745 		powersave_nap = 0;
746 }
747 
748 static int __init smp_core99_probe(void)
749 {
750 	struct device_node *cpus;
751 	int ncpus = 0;
752 
753 	if (ppc_md.progress) ppc_md.progress("smp_core99_probe", 0x345);
754 
755 	/* Count CPUs in the device-tree */
756        	for (cpus = NULL; (cpus = of_find_node_by_type(cpus, "cpu")) != NULL;)
757 	       	++ncpus;
758 
759 	printk(KERN_INFO "PowerMac SMP probe found %d cpus\n", ncpus);
760 
761 	/* Nothing more to do if less than 2 of them */
762 	if (ncpus <= 1)
763 		return 1;
764 
765 	/* We need to perform some early initialisations before we can start
766 	 * setting up SMP as we are running before initcalls
767 	 */
768 	pmac_pfunc_base_install();
769 	pmac_i2c_init();
770 
771 	/* Setup various bits like timebase sync method, ability to nap, ... */
772 	smp_core99_setup(ncpus);
773 
774 	/* Install IPIs */
775 	mpic_request_ipis();
776 
777 	/* Collect l2cr and l3cr values from CPU 0 */
778 	core99_init_caches(0);
779 
780 	return ncpus;
781 }
782 
783 static void __devinit smp_core99_kick_cpu(int nr)
784 {
785 	unsigned int save_vector;
786 	unsigned long target, flags;
787 	volatile unsigned int *vector
788 		 = ((volatile unsigned int *)(KERNELBASE+0x100));
789 
790 	if (nr < 0 || nr > 3)
791 		return;
792 
793 	if (ppc_md.progress)
794 		ppc_md.progress("smp_core99_kick_cpu", 0x346);
795 
796 	local_irq_save(flags);
797 	local_irq_disable();
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