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 <linux/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#include "pmac.h"
57
58#undef DEBUG
59
60#ifdef DEBUG
61#define DBG(fmt...) udbg_printf(fmt)
62#else
63#define DBG(fmt...)
64#endif
65
66extern void __secondary_start_pmac_0(void);
67extern int pmac_pfunc_base_install(void);
68
69static void (*pmac_tb_freeze)(int freeze);
70static u64 timebase;
71static int tb_req;
72
73#ifdef CONFIG_PPC_PMAC32_PSURGE
74
75/*
76 * Powersurge (old powermac SMP) support.
77 */
78
79/* Addresses for powersurge registers */
80#define HAMMERHEAD_BASE		0xf8000000
81#define HHEAD_CONFIG		0x90
82#define HHEAD_SEC_INTR		0xc0
83
84/* register for interrupting the primary processor on the powersurge */
85/* N.B. this is actually the ethernet ROM! */
86#define PSURGE_PRI_INTR		0xf3019000
87
88/* register for storing the start address for the secondary processor */
89/* N.B. this is the PCI config space address register for the 1st bridge */
90#define PSURGE_START		0xf2800000
91
92/* Daystar/XLR8 4-CPU card */
93#define PSURGE_QUAD_REG_ADDR	0xf8800000
94
95#define PSURGE_QUAD_IRQ_SET	0
96#define PSURGE_QUAD_IRQ_CLR	1
97#define PSURGE_QUAD_IRQ_PRIMARY	2
98#define PSURGE_QUAD_CKSTOP_CTL	3
99#define PSURGE_QUAD_PRIMARY_ARB	4
100#define PSURGE_QUAD_BOARD_ID	6
101#define PSURGE_QUAD_WHICH_CPU	7
102#define PSURGE_QUAD_CKSTOP_RDBK	8
103#define PSURGE_QUAD_RESET_CTL	11
104
105#define PSURGE_QUAD_OUT(r, v)	(out_8(quad_base + ((r) << 4) + 4, (v)))
106#define PSURGE_QUAD_IN(r)	(in_8(quad_base + ((r) << 4) + 4) & 0x0f)
107#define PSURGE_QUAD_BIS(r, v)	(PSURGE_QUAD_OUT((r), PSURGE_QUAD_IN(r) | (v)))
108#define PSURGE_QUAD_BIC(r, v)	(PSURGE_QUAD_OUT((r), PSURGE_QUAD_IN(r) & ~(v)))
109
110/* virtual addresses for the above */
111static volatile u8 __iomem *hhead_base;
112static volatile u8 __iomem *quad_base;
113static volatile u32 __iomem *psurge_pri_intr;
114static volatile u8 __iomem *psurge_sec_intr;
115static volatile u32 __iomem *psurge_start;
116
117/* values for psurge_type */
118#define PSURGE_NONE		-1
119#define PSURGE_DUAL		0
120#define PSURGE_QUAD_OKEE	1
121#define PSURGE_QUAD_COTTON	2
122#define PSURGE_QUAD_ICEGRASS	3
123
124/* what sort of powersurge board we have */
125static int psurge_type = PSURGE_NONE;
126
127/* irq for secondary cpus to report */
128static struct irq_domain *psurge_host;
129int psurge_secondary_virq;
130
131/*
132 * Set and clear IPIs for powersurge.
133 */
134static inline void psurge_set_ipi(int cpu)
135{
136	if (psurge_type == PSURGE_NONE)
137		return;
138	if (cpu == 0)
139		in_be32(psurge_pri_intr);
140	else if (psurge_type == PSURGE_DUAL)
141		out_8(psurge_sec_intr, 0);
142	else
143		PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_SET, 1 << cpu);
144}
145
146static inline void psurge_clr_ipi(int cpu)
147{
148	if (cpu > 0) {
149		switch(psurge_type) {
150		case PSURGE_DUAL:
151			out_8(psurge_sec_intr, ~0);
152		case PSURGE_NONE:
153			break;
154		default:
155			PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_CLR, 1 << cpu);
156		}
157	}
158}
159
160/*
161 * On powersurge (old SMP powermac architecture) we don't have
162 * separate IPIs for separate messages like openpic does.  Instead
163 * use the generic demux helpers
164 *  -- paulus.
165 */
166static irqreturn_t psurge_ipi_intr(int irq, void *d)
167{
168	psurge_clr_ipi(smp_processor_id());
169	smp_ipi_demux();
170
171	return IRQ_HANDLED;
172}
173
174static void smp_psurge_cause_ipi(int cpu, unsigned long data)
175{
176	psurge_set_ipi(cpu);
177}
178
179static int psurge_host_map(struct irq_domain *h, unsigned int virq,
180			 irq_hw_number_t hw)
181{
182	irq_set_chip_and_handler(virq, &dummy_irq_chip, handle_percpu_irq);
183
184	return 0;
185}
186
187static const struct irq_domain_ops psurge_host_ops = {
188	.map	= psurge_host_map,
189};
190
191static int psurge_secondary_ipi_init(void)
192{
193	int rc = -ENOMEM;
194
195	psurge_host = irq_domain_add_nomap(NULL, ~0, &psurge_host_ops, NULL);
196
197	if (psurge_host)
198		psurge_secondary_virq = irq_create_direct_mapping(psurge_host);
199
200	if (psurge_secondary_virq)
201		rc = request_irq(psurge_secondary_virq, psurge_ipi_intr,
202			IRQF_PERCPU | IRQF_NO_THREAD, "IPI", NULL);
203
204	if (rc)
205		pr_err("Failed to setup secondary cpu IPI\n");
206
207	return rc;
208}
209
210/*
211 * Determine a quad card presence. We read the board ID register, we
212 * force the data bus to change to something else, and we read it again.
213 * It it's stable, then the register probably exist (ugh !)
214 */
215static int __init psurge_quad_probe(void)
216{
217	int type;
218	unsigned int i;
219
220	type = PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID);
221	if (type < PSURGE_QUAD_OKEE || type > PSURGE_QUAD_ICEGRASS
222	    || type != PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID))
223		return PSURGE_DUAL;
224
225	/* looks OK, try a slightly more rigorous test */
226	/* bogus is not necessarily cacheline-aligned,
227	   though I don't suppose that really matters.  -- paulus */
228	for (i = 0; i < 100; i++) {
229		volatile u32 bogus[8];
230		bogus[(0+i)%8] = 0x00000000;
231		bogus[(1+i)%8] = 0x55555555;
232		bogus[(2+i)%8] = 0xFFFFFFFF;
233		bogus[(3+i)%8] = 0xAAAAAAAA;
234		bogus[(4+i)%8] = 0x33333333;
235		bogus[(5+i)%8] = 0xCCCCCCCC;
236		bogus[(6+i)%8] = 0xCCCCCCCC;
237		bogus[(7+i)%8] = 0x33333333;
238		wmb();
239		asm volatile("dcbf 0,%0" : : "r" (bogus) : "memory");
240		mb();
241		if (type != PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID))
242			return PSURGE_DUAL;
243	}
244	return type;
245}
246
247static void __init psurge_quad_init(void)
248{
249	int procbits;
250
251	if (ppc_md.progress) ppc_md.progress("psurge_quad_init", 0x351);
252	procbits = ~PSURGE_QUAD_IN(PSURGE_QUAD_WHICH_CPU);
253	if (psurge_type == PSURGE_QUAD_ICEGRASS)
254		PSURGE_QUAD_BIS(PSURGE_QUAD_RESET_CTL, procbits);
255	else
256		PSURGE_QUAD_BIC(PSURGE_QUAD_CKSTOP_CTL, procbits);
257	mdelay(33);
258	out_8(psurge_sec_intr, ~0);
259	PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_CLR, procbits);
260	PSURGE_QUAD_BIS(PSURGE_QUAD_RESET_CTL, procbits);
261	if (psurge_type != PSURGE_QUAD_ICEGRASS)
262		PSURGE_QUAD_BIS(PSURGE_QUAD_CKSTOP_CTL, procbits);
263	PSURGE_QUAD_BIC(PSURGE_QUAD_PRIMARY_ARB, procbits);
264	mdelay(33);
265	PSURGE_QUAD_BIC(PSURGE_QUAD_RESET_CTL, procbits);
266	mdelay(33);
267	PSURGE_QUAD_BIS(PSURGE_QUAD_PRIMARY_ARB, procbits);
268	mdelay(33);
269}
270
271static void __init smp_psurge_probe(void)
272{
273	int i, ncpus;
274	struct device_node *dn;
275
276	/* We don't do SMP on the PPC601 -- paulus */
277	if (PVR_VER(mfspr(SPRN_PVR)) == 1)
278		return;
279
280	/*
281	 * The powersurge cpu board can be used in the generation
282	 * of powermacs that have a socket for an upgradeable cpu card,
283	 * including the 7500, 8500, 9500, 9600.
284	 * The device tree doesn't tell you if you have 2 cpus because
285	 * OF doesn't know anything about the 2nd processor.
286	 * Instead we look for magic bits in magic registers,
287	 * in the hammerhead memory controller in the case of the
288	 * dual-cpu powersurge board.  -- paulus.
289	 */
290	dn = of_find_node_by_name(NULL, "hammerhead");
291	if (dn == NULL)
292		return;
293	of_node_put(dn);
294
295	hhead_base = ioremap(HAMMERHEAD_BASE, 0x800);
296	quad_base = ioremap(PSURGE_QUAD_REG_ADDR, 1024);
297	psurge_sec_intr = hhead_base + HHEAD_SEC_INTR;
298
299	psurge_type = psurge_quad_probe();
300	if (psurge_type != PSURGE_DUAL) {
301		psurge_quad_init();
302		/* All released cards using this HW design have 4 CPUs */
303		ncpus = 4;
304		/* No sure how timebase sync works on those, let's use SW */
305		smp_ops->give_timebase = smp_generic_give_timebase;
306		smp_ops->take_timebase = smp_generic_take_timebase;
307	} else {
308		iounmap(quad_base);
309		if ((in_8(hhead_base + HHEAD_CONFIG) & 0x02) == 0) {
310			/* not a dual-cpu card */
311			iounmap(hhead_base);
312			psurge_type = PSURGE_NONE;
313			return;
314		}
315		ncpus = 2;
316	}
317
318	if (psurge_secondary_ipi_init())
319		return;
320
321	psurge_start = ioremap(PSURGE_START, 4);
322	psurge_pri_intr = ioremap(PSURGE_PRI_INTR, 4);
323
324	/* This is necessary because OF doesn't know about the
325	 * secondary cpu(s), and thus there aren't nodes in the
326	 * device tree for them, and smp_setup_cpu_maps hasn't
327	 * set their bits in cpu_present_mask.
328	 */
329	if (ncpus > NR_CPUS)
330		ncpus = NR_CPUS;
331	for (i = 1; i < ncpus ; ++i)
332		set_cpu_present(i, true);
333
334	if (ppc_md.progress) ppc_md.progress("smp_psurge_probe - done", 0x352);
335}
336
337static int __init smp_psurge_kick_cpu(int nr)
338{
339	unsigned long start = __pa(__secondary_start_pmac_0) + nr * 8;
340	unsigned long a, flags;
341	int i, j;
342
343	/* Defining this here is evil ... but I prefer hiding that
344	 * crap to avoid giving people ideas that they can do the
345	 * same.
346	 */
347	extern volatile unsigned int cpu_callin_map[NR_CPUS];
348
349	/* may need to flush here if secondary bats aren't setup */
350	for (a = KERNELBASE; a < KERNELBASE + 0x800000; a += 32)
351		asm volatile("dcbf 0,%0" : : "r" (a) : "memory");
352	asm volatile("sync");
353
354	if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu", 0x353);
355
356	/* This is going to freeze the timeebase, we disable interrupts */
357	local_irq_save(flags);
358
359	out_be32(psurge_start, start);
360	mb();
361
362	psurge_set_ipi(nr);
363
364	/*
365	 * We can't use udelay here because the timebase is now frozen.
366	 */
367	for (i = 0; i < 2000; ++i)
368		asm volatile("nop" : : : "memory");
369	psurge_clr_ipi(nr);
370
371	/*
372	 * Also, because the timebase is frozen, we must not return to the
373	 * caller which will try to do udelay's etc... Instead, we wait -here-
374	 * for the CPU to callin.
375	 */
376	for (i = 0; i < 100000 && !cpu_callin_map[nr]; ++i) {
377		for (j = 1; j < 10000; j++)
378			asm volatile("nop" : : : "memory");
379		asm volatile("sync" : : : "memory");
380	}
381	if (!cpu_callin_map[nr])
382		goto stuck;
383
384	/* And we do the TB sync here too for standard dual CPU cards */
385	if (psurge_type == PSURGE_DUAL) {
386		while(!tb_req)
387			barrier();
388		tb_req = 0;
389		mb();
390		timebase = get_tb();
391		mb();
392		while (timebase)
393			barrier();
394		mb();
395	}
396 stuck:
397	/* now interrupt the secondary, restarting both TBs */
398	if (psurge_type == PSURGE_DUAL)
399		psurge_set_ipi(1);
400
401	if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu - done", 0x354);
402
403	return 0;
404}
405
406static struct irqaction psurge_irqaction = {
407	.handler = psurge_ipi_intr,
408	.flags = IRQF_PERCPU | IRQF_NO_THREAD,
409	.name = "primary IPI",
410};
411
412static void __init smp_psurge_setup_cpu(int cpu_nr)
413{
414	if (cpu_nr != 0 || !psurge_start)
415		return;
416
417	/* reset the entry point so if we get another intr we won't
418	 * try to startup again */
419	out_be32(psurge_start, 0x100);
420	if (setup_irq(irq_create_mapping(NULL, 30), &psurge_irqaction))
421		printk(KERN_ERR "Couldn't get primary IPI interrupt");
422}
423
424void __init smp_psurge_take_timebase(void)
425{
426	if (psurge_type != PSURGE_DUAL)
427		return;
428
429	tb_req = 1;
430	mb();
431	while (!timebase)
432		barrier();
433	mb();
434	set_tb(timebase >> 32, timebase & 0xffffffff);
435	timebase = 0;
436	mb();
437	set_dec(tb_ticks_per_jiffy/2);
438}
439
440void __init smp_psurge_give_timebase(void)
441{
442	/* Nothing to do here */
443}
444
445/* PowerSurge-style Macs */
446struct smp_ops_t psurge_smp_ops = {
447	.message_pass	= NULL,	/* Use smp_muxed_ipi_message_pass */
448	.cause_ipi	= smp_psurge_cause_ipi,
449	.probe		= smp_psurge_probe,
450	.kick_cpu	= smp_psurge_kick_cpu,
451	.setup_cpu	= smp_psurge_setup_cpu,
452	.give_timebase	= smp_psurge_give_timebase,
453	.take_timebase	= smp_psurge_take_timebase,
454};
455#endif /* CONFIG_PPC_PMAC32_PSURGE */
456
457/*
458 * Core 99 and later support
459 */
460
461
462static void smp_core99_give_timebase(void)
463{
464	unsigned long flags;
465
466	local_irq_save(flags);
467
468	while(!tb_req)
469		barrier();
470	tb_req = 0;
471	(*pmac_tb_freeze)(1);
472	mb();
473	timebase = get_tb();
474	mb();
475	while (timebase)
476		barrier();
477	mb();
478	(*pmac_tb_freeze)(0);
479	mb();
480
481	local_irq_restore(flags);
482}
483
484
485static void smp_core99_take_timebase(void)
486{
487	unsigned long flags;
488
489	local_irq_save(flags);
490
491	tb_req = 1;
492	mb();
493	while (!timebase)
494		barrier();
495	mb();
496	set_tb(timebase >> 32, timebase & 0xffffffff);
497	timebase = 0;
498	mb();
499
500	local_irq_restore(flags);
501}
502
503#ifdef CONFIG_PPC64
504/*
505 * G5s enable/disable the timebase via an i2c-connected clock chip.
506 */
507static struct pmac_i2c_bus *pmac_tb_clock_chip_host;
508static u8 pmac_tb_pulsar_addr;
509
510static void smp_core99_cypress_tb_freeze(int freeze)
511{
512	u8 data;
513	int rc;
514
515	/* Strangely, the device-tree says address is 0xd2, but darwin
516	 * accesses 0xd0 ...
517	 */
518	pmac_i2c_setmode(pmac_tb_clock_chip_host,
519			 pmac_i2c_mode_combined);
520	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
521			   0xd0 | pmac_i2c_read,
522			   1, 0x81, &data, 1);
523	if (rc != 0)
524		goto bail;
525
526	data = (data & 0xf3) | (freeze ? 0x00 : 0x0c);
527
528       	pmac_i2c_setmode(pmac_tb_clock_chip_host, pmac_i2c_mode_stdsub);
529	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
530			   0xd0 | pmac_i2c_write,
531			   1, 0x81, &data, 1);
532
533 bail:
534	if (rc != 0) {
535		printk("Cypress Timebase %s rc: %d\n",
536		       freeze ? "freeze" : "unfreeze", rc);
537		panic("Timebase freeze failed !\n");
538	}
539}
540
541
542static void smp_core99_pulsar_tb_freeze(int freeze)
543{
544	u8 data;
545	int rc;
546
547	pmac_i2c_setmode(pmac_tb_clock_chip_host,
548			 pmac_i2c_mode_combined);
549	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
550			   pmac_tb_pulsar_addr | pmac_i2c_read,
551			   1, 0x2e, &data, 1);
552	if (rc != 0)
553		goto bail;
554
555	data = (data & 0x88) | (freeze ? 0x11 : 0x22);
556
557	pmac_i2c_setmode(pmac_tb_clock_chip_host, pmac_i2c_mode_stdsub);
558	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
559			   pmac_tb_pulsar_addr | pmac_i2c_write,
560			   1, 0x2e, &data, 1);
561 bail:
562	if (rc != 0) {
563		printk(KERN_ERR "Pulsar Timebase %s rc: %d\n",
564		       freeze ? "freeze" : "unfreeze", rc);
565		panic("Timebase freeze failed !\n");
566	}
567}
568
569static void __init smp_core99_setup_i2c_hwsync(int ncpus)
570{
571	struct device_node *cc = NULL;
572	struct device_node *p;
573	const char *name = NULL;
574	const u32 *reg;
575	int ok;
576
577	/* Look for the clock chip */
578	for_each_node_by_name(cc, "i2c-hwclock") {
579		p = of_get_parent(cc);
580		ok = p && of_device_is_compatible(p, "uni-n-i2c");
581		of_node_put(p);
582		if (!ok)
583			continue;
584
585		pmac_tb_clock_chip_host = pmac_i2c_find_bus(cc);
586		if (pmac_tb_clock_chip_host == NULL)
587			continue;
588		reg = of_get_property(cc, "reg", NULL);
589		if (reg == NULL)
590			continue;
591		switch (*reg) {
592		case 0xd2:
593			if (of_device_is_compatible(cc,"pulsar-legacy-slewing")) {
594				pmac_tb_freeze = smp_core99_pulsar_tb_freeze;
595				pmac_tb_pulsar_addr = 0xd2;
596				name = "Pulsar";
597			} else if (of_device_is_compatible(cc, "cy28508")) {
598				pmac_tb_freeze = smp_core99_cypress_tb_freeze;
599				name = "Cypress";
600			}
601			break;
602		case 0xd4:
603			pmac_tb_freeze = smp_core99_pulsar_tb_freeze;
604			pmac_tb_pulsar_addr = 0xd4;
605			name = "Pulsar";
606			break;
607		}
608		if (pmac_tb_freeze != NULL)
609			break;
610	}
611	if (pmac_tb_freeze != NULL) {
612		/* Open i2c bus for synchronous access */
613		if (pmac_i2c_open(pmac_tb_clock_chip_host, 1)) {
614			printk(KERN_ERR "Failed top open i2c bus for clock"
615			       " sync, fallback to software sync !\n");
616			goto no_i2c_sync;
617		}
618		printk(KERN_INFO "Processor timebase sync using %s i2c clock\n",
619		       name);
620		return;
621	}
622 no_i2c_sync:
623	pmac_tb_freeze = NULL;
624	pmac_tb_clock_chip_host = NULL;
625}
626
627
628
629/*
630 * Newer G5s uses a platform function
631 */
632
633static void smp_core99_pfunc_tb_freeze(int freeze)
634{
635	struct device_node *cpus;
636	struct pmf_args args;
637
638	cpus = of_find_node_by_path("/cpus");
639	BUG_ON(cpus == NULL);
640	args.count = 1;
641	args.u[0].v = !freeze;
642	pmf_call_function(cpus, "cpu-timebase", &args);
643	of_node_put(cpus);
644}
645
646#else /* CONFIG_PPC64 */
647
648/*
649 * SMP G4 use a GPIO to enable/disable the timebase.
650 */
651
652static unsigned int core99_tb_gpio;	/* Timebase freeze GPIO */
653
654static void smp_core99_gpio_tb_freeze(int freeze)
655{
656	if (freeze)
657		pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 4);
658	else
659		pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 0);
660	pmac_call_feature(PMAC_FTR_READ_GPIO, NULL, core99_tb_gpio, 0);
661}
662
663
664#endif /* !CONFIG_PPC64 */
665
666/* L2 and L3 cache settings to pass from CPU0 to CPU1 on G4 cpus */
667volatile static long int core99_l2_cache;
668volatile static long int core99_l3_cache;
669
670static void core99_init_caches(int cpu)
671{
672#ifndef CONFIG_PPC64
673	if (!cpu_has_feature(CPU_FTR_L2CR))
674		return;
675
676	if (cpu == 0) {
677		core99_l2_cache = _get_L2CR();
678		printk("CPU0: L2CR is %lx\n", core99_l2_cache);
679	} else {
680		printk("CPU%d: L2CR was %lx\n", cpu, _get_L2CR());
681		_set_L2CR(0);
682		_set_L2CR(core99_l2_cache);
683		printk("CPU%d: L2CR set to %lx\n", cpu, core99_l2_cache);
684	}
685
686	if (!cpu_has_feature(CPU_FTR_L3CR))
687		return;
688
689	if (cpu == 0){
690		core99_l3_cache = _get_L3CR();
691		printk("CPU0: L3CR is %lx\n", core99_l3_cache);
692	} else {
693		printk("CPU%d: L3CR was %lx\n", cpu, _get_L3CR());
694		_set_L3CR(0);
695		_set_L3CR(core99_l3_cache);
696		printk("CPU%d: L3CR set to %lx\n", cpu, core99_l3_cache);
697	}
698#endif /* !CONFIG_PPC64 */
699}
700
701static void __init smp_core99_setup(int ncpus)
702{
703#ifdef CONFIG_PPC64
704
705	/* i2c based HW sync on some G5s */
706	if (of_machine_is_compatible("PowerMac7,2") ||
707	    of_machine_is_compatible("PowerMac7,3") ||
708	    of_machine_is_compatible("RackMac3,1"))
709		smp_core99_setup_i2c_hwsync(ncpus);
710
711	/* pfunc based HW sync on recent G5s */
712	if (pmac_tb_freeze == NULL) {
713		struct device_node *cpus =
714			of_find_node_by_path("/cpus");
715		if (cpus &&
716		    of_get_property(cpus, "platform-cpu-timebase", NULL)) {
717			pmac_tb_freeze = smp_core99_pfunc_tb_freeze;
718			printk(KERN_INFO "Processor timebase sync using"
719			       " platform function\n");
720		}
721	}
722
723#else /* CONFIG_PPC64 */
724
725	/* GPIO based HW sync on ppc32 Core99 */
726	if (pmac_tb_freeze == NULL && !of_machine_is_compatible("MacRISC4")) {
727		struct device_node *cpu;
728		const u32 *tbprop = NULL;
729
730		core99_tb_gpio = KL_GPIO_TB_ENABLE;	/* default value */
731		cpu = of_find_node_by_type(NULL, "cpu");
732		if (cpu != NULL) {
733			tbprop = of_get_property(cpu, "timebase-enable", NULL);
734			if (tbprop)
735				core99_tb_gpio = *tbprop;
736			of_node_put(cpu);
737		}
738		pmac_tb_freeze = smp_core99_gpio_tb_freeze;
739		printk(KERN_INFO "Processor timebase sync using"
740		       " GPIO 0x%02x\n", core99_tb_gpio);
741	}
742
743#endif /* CONFIG_PPC64 */
744
745	/* No timebase sync, fallback to software */
746	if (pmac_tb_freeze == NULL) {
747		smp_ops->give_timebase = smp_generic_give_timebase;
748		smp_ops->take_timebase = smp_generic_take_timebase;
749		printk(KERN_INFO "Processor timebase sync using software\n");
750	}
751
752#ifndef CONFIG_PPC64
753	{
754		int i;
755
756		/* XXX should get this from reg properties */
757		for (i = 1; i < ncpus; ++i)
758			set_hard_smp_processor_id(i, i);
759	}
760#endif
761
762	/* 32 bits SMP can't NAP */
763	if (!of_machine_is_compatible("MacRISC4"))
764		powersave_nap = 0;
765}
766
767static void __init smp_core99_probe(void)
768{
769	struct device_node *cpus;
770	int ncpus = 0;
771
772	if (ppc_md.progress) ppc_md.progress("smp_core99_probe", 0x345);
773
774	/* Count CPUs in the device-tree */
775       	for (cpus = NULL; (cpus = of_find_node_by_type(cpus, "cpu")) != NULL;)
776	       	++ncpus;
777
778	printk(KERN_INFO "PowerMac SMP probe found %d cpus\n", ncpus);
779
780	/* Nothing more to do if less than 2 of them */
781	if (ncpus <= 1)
782		return;
783
784	/* We need to perform some early initialisations before we can start
785	 * setting up SMP as we are running before initcalls
786	 */
787	pmac_pfunc_base_install();
788	pmac_i2c_init();
789
790	/* Setup various bits like timebase sync method, ability to nap, ... */
791	smp_core99_setup(ncpus);
792
793	/* Install IPIs */
794	mpic_request_ipis();
795
796	/* Collect l2cr and l3cr values from CPU 0 */
797	core99_init_caches(0);
798}
799
800static int smp_core99_kick_cpu(int nr)
801{
802	unsigned int save_vector;
803	unsigned long target, flags;
804	unsigned int *vector = (unsigned int *)(PAGE_OFFSET+0x100);
805
806	if (nr < 0 || nr > 3)
807		return -ENOENT;
808
809	if (ppc_md.progress)
810		ppc_md.progress("smp_core99_kick_cpu", 0x346);
811
812	local_irq_save(flags);
813
814	/* Save reset vector */
815	save_vector = *vector;
816
817	/* Setup fake reset vector that does
818	 *   b __secondary_start_pmac_0 + nr*8
819	 */
820	target = (unsigned long) __secondary_start_pmac_0 + nr * 8;
821	patch_branch(vector, target, BRANCH_SET_LINK);
822
823	/* Put some life in our friend */
824	pmac_call_feature(PMAC_FTR_RESET_CPU, NULL, nr, 0);
825
826	/* FIXME: We wait a bit for the CPU to take the exception, I should
827	 * instead wait for the entry code to set something for me. Well,
828	 * ideally, all that crap will be done in prom.c and the CPU left
829	 * in a RAM-based wait loop like CHRP.
830	 */
831	mdelay(1);
832
833	/* Restore our exception vector */
834	*vector = save_vector;
835	flush_icache_range((unsigned long) vector, (unsigned long) vector + 4);
836
837	local_irq_restore(flags);
838	if (ppc_md.progress) ppc_md.progress("smp_core99_kick_cpu done", 0x347);
839
840	return 0;
841}
842
843static void smp_core99_setup_cpu(int cpu_nr)
844{
845	/* Setup L2/L3 */
846	if (cpu_nr != 0)
847		core99_init_caches(cpu_nr);
848
849	/* Setup openpic */
850	mpic_setup_this_cpu();
851}
852
853#ifdef CONFIG_PPC64
854#ifdef CONFIG_HOTPLUG_CPU
855static int smp_core99_cpu_notify(struct notifier_block *self,
856				 unsigned long action, void *hcpu)
857{
858	int rc;
859
860	switch(action) {
861	case CPU_UP_PREPARE:
862	case CPU_UP_PREPARE_FROZEN:
863		/* Open i2c bus if it was used for tb sync */
864		if (pmac_tb_clock_chip_host) {
865			rc = pmac_i2c_open(pmac_tb_clock_chip_host, 1);
866			if (rc) {
867				pr_err("Failed to open i2c bus for time sync\n");
868				return notifier_from_errno(rc);
869			}
870		}
871		break;
872	case CPU_ONLINE:
873	case CPU_UP_CANCELED:
874		/* Close i2c bus if it was used for tb sync */
875		if (pmac_tb_clock_chip_host)
876			pmac_i2c_close(pmac_tb_clock_chip_host);
877		break;
878	default:
879		break;
880	}
881	return NOTIFY_OK;
882}
883
884static struct notifier_block smp_core99_cpu_nb = {
885	.notifier_call	= smp_core99_cpu_notify,
886};
887#endif /* CONFIG_HOTPLUG_CPU */
888
889static void __init smp_core99_bringup_done(void)
890{
891	extern void g5_phy_disable_cpu1(void);
892
893	/* Close i2c bus if it was used for tb sync */
894	if (pmac_tb_clock_chip_host)
895		pmac_i2c_close(pmac_tb_clock_chip_host);
896
897	/* If we didn't start the second CPU, we must take
898	 * it off the bus.
899	 */
900	if (of_machine_is_compatible("MacRISC4") &&
901	    num_online_cpus() < 2) {
902		set_cpu_present(1, false);
903		g5_phy_disable_cpu1();
904	}
905#ifdef CONFIG_HOTPLUG_CPU
906	register_cpu_notifier(&smp_core99_cpu_nb);
907#endif
908
909	if (ppc_md.progress)
910		ppc_md.progress("smp_core99_bringup_done", 0x349);
911}
912#endif /* CONFIG_PPC64 */
913
914#ifdef CONFIG_HOTPLUG_CPU
915
916static int smp_core99_cpu_disable(void)
917{
918	int rc = generic_cpu_disable();
919	if (rc)
920		return rc;
921
922	mpic_cpu_set_priority(0xf);
923
924	return 0;
925}
926
927#ifdef CONFIG_PPC32
928
929static void pmac_cpu_die(void)
930{
931	int cpu = smp_processor_id();
932
933	local_irq_disable();
934	idle_task_exit();
935	pr_debug("CPU%d offline\n", cpu);
936	generic_set_cpu_dead(cpu);
937	smp_wmb();
938	mb();
939	low_cpu_die();
940}
941
942#else /* CONFIG_PPC32 */
943
944static void pmac_cpu_die(void)
945{
946	int cpu = smp_processor_id();
947
948	local_irq_disable();
949	idle_task_exit();
950
951	/*
952	 * turn off as much as possible, we'll be
953	 * kicked out as this will only be invoked
954	 * on core99 platforms for now ...
955	 */
956
957	printk(KERN_INFO "CPU#%d offline\n", cpu);
958	generic_set_cpu_dead(cpu);
959	smp_wmb();
960
961	/*
962	 * Re-enable interrupts. The NAP code needs to enable them
963	 * anyways, do it now so we deal with the case where one already
964	 * happened while soft-disabled.
965	 * We shouldn't get any external interrupts, only decrementer, and the
966	 * decrementer handler is safe for use on offline CPUs
967	 */
968	local_irq_enable();
969
970	while (1) {
971		/* let's not take timer interrupts too often ... */
972		set_dec(0x7fffffff);
973
974		/* Enter NAP mode */
975		power4_idle();
976	}
977}
978
979#endif /* else CONFIG_PPC32 */
980#endif /* CONFIG_HOTPLUG_CPU */
981
982/* Core99 Macs (dual G4s and G5s) */
983struct smp_ops_t core99_smp_ops = {
984	.message_pass	= smp_mpic_message_pass,
985	.probe		= smp_core99_probe,
986#ifdef CONFIG_PPC64
987	.bringup_done	= smp_core99_bringup_done,
988#endif
989	.kick_cpu	= smp_core99_kick_cpu,
990	.setup_cpu	= smp_core99_setup_cpu,
991	.give_timebase	= smp_core99_give_timebase,
992	.take_timebase	= smp_core99_take_timebase,
993#if defined(CONFIG_HOTPLUG_CPU)
994	.cpu_disable	= smp_core99_cpu_disable,
995	.cpu_die	= generic_cpu_die,
996#endif
997};
998
999void __init pmac_setup_smp(void)
1000{
1001	struct device_node *np;
1002
1003	/* Check for Core99 */
1004	np = of_find_node_by_name(NULL, "uni-n");
1005	if (!np)
1006		np = of_find_node_by_name(NULL, "u3");
1007	if (!np)
1008		np = of_find_node_by_name(NULL, "u4");
1009	if (np) {
1010		of_node_put(np);
1011		smp_ops = &core99_smp_ops;
1012	}
1013#ifdef CONFIG_PPC_PMAC32_PSURGE
1014	else {
1015		/* We have to set bits in cpu_possible_mask here since the
1016		 * secondary CPU(s) aren't in the device tree. Various
1017		 * things won't be initialized for CPUs not in the possible
1018		 * map, so we really need to fix it up here.
1019		 */
1020		int cpu;
1021
1022		for (cpu = 1; cpu < 4 && cpu < NR_CPUS; ++cpu)
1023			set_cpu_possible(cpu, true);
1024		smp_ops = &psurge_smp_ops;
1025	}
1026#endif /* CONFIG_PPC_PMAC32_PSURGE */
1027
1028#ifdef CONFIG_HOTPLUG_CPU
1029	ppc_md.cpu_die = pmac_cpu_die;
1030#endif
1031}
1032
1033
1034