PageRenderTime 50ms CodeModel.GetById 17ms app.highlight 27ms RepoModel.GetById 1ms app.codeStats 0ms

/arch/powerpc/platforms/powermac/smp.c

https://bitbucket.org/abioy/linux
C | 960 lines | 655 code | 151 blank | 154 comment | 141 complexity | 3108118daf0d6512f7e2b7a057cac110 MD5 | raw file
Possible License(s): CC-BY-SA-3.0, GPL-2.0, LGPL-2.0, AGPL-1.0
  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#undef DEBUG
 57
 58#ifdef DEBUG
 59#define DBG(fmt...) udbg_printf(fmt)
 60#else
 61#define DBG(fmt...)
 62#endif
 63
 64extern void __secondary_start_pmac_0(void);
 65extern int pmac_pfunc_base_install(void);
 66
 67static void (*pmac_tb_freeze)(int freeze);
 68static u64 timebase;
 69static int tb_req;
 70
 71#ifdef CONFIG_PPC32
 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 */
109static volatile u8 __iomem *hhead_base;
110static volatile u8 __iomem *quad_base;
111static volatile u32 __iomem *psurge_pri_intr;
112static volatile u8 __iomem *psurge_sec_intr;
113static 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 */
123static int psurge_type = PSURGE_NONE;
124
125/*
126 * Set and clear IPIs for powersurge.
127 */
128static 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
140static 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 */
162static unsigned long psurge_smp_message[NR_CPUS];
163
164void psurge_smp_message_recv(void)
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);
179}
180
181irqreturn_t psurge_primary_intr(int irq, void *d)
182{
183	psurge_smp_message_recv();
184	return IRQ_HANDLED;
185}
186
187static 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_each_online_cpu(i) {
195		if (target == MSG_ALL
196		    || (target == MSG_ALL_BUT_SELF && i != smp_processor_id())
197		    || target == i) {
198			set_bit(msg, &psurge_smp_message[i]);
199			psurge_set_ipi(i);
200		}
201	}
202}
203
204/*
205 * Determine a quad card presence. We read the board ID register, we
206 * force the data bus to change to something else, and we read it again.
207 * It it's stable, then the register probably exist (ugh !)
208 */
209static int __init psurge_quad_probe(void)
210{
211	int type;
212	unsigned int i;
213
214	type = PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID);
215	if (type < PSURGE_QUAD_OKEE || type > PSURGE_QUAD_ICEGRASS
216	    || type != PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID))
217		return PSURGE_DUAL;
218
219	/* looks OK, try a slightly more rigorous test */
220	/* bogus is not necessarily cacheline-aligned,
221	   though I don't suppose that really matters.  -- paulus */
222	for (i = 0; i < 100; i++) {
223		volatile u32 bogus[8];
224		bogus[(0+i)%8] = 0x00000000;
225		bogus[(1+i)%8] = 0x55555555;
226		bogus[(2+i)%8] = 0xFFFFFFFF;
227		bogus[(3+i)%8] = 0xAAAAAAAA;
228		bogus[(4+i)%8] = 0x33333333;
229		bogus[(5+i)%8] = 0xCCCCCCCC;
230		bogus[(6+i)%8] = 0xCCCCCCCC;
231		bogus[(7+i)%8] = 0x33333333;
232		wmb();
233		asm volatile("dcbf 0,%0" : : "r" (bogus) : "memory");
234		mb();
235		if (type != PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID))
236			return PSURGE_DUAL;
237	}
238	return type;
239}
240
241static void __init psurge_quad_init(void)
242{
243	int procbits;
244
245	if (ppc_md.progress) ppc_md.progress("psurge_quad_init", 0x351);
246	procbits = ~PSURGE_QUAD_IN(PSURGE_QUAD_WHICH_CPU);
247	if (psurge_type == PSURGE_QUAD_ICEGRASS)
248		PSURGE_QUAD_BIS(PSURGE_QUAD_RESET_CTL, procbits);
249	else
250		PSURGE_QUAD_BIC(PSURGE_QUAD_CKSTOP_CTL, procbits);
251	mdelay(33);
252	out_8(psurge_sec_intr, ~0);
253	PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_CLR, procbits);
254	PSURGE_QUAD_BIS(PSURGE_QUAD_RESET_CTL, procbits);
255	if (psurge_type != PSURGE_QUAD_ICEGRASS)
256		PSURGE_QUAD_BIS(PSURGE_QUAD_CKSTOP_CTL, procbits);
257	PSURGE_QUAD_BIC(PSURGE_QUAD_PRIMARY_ARB, procbits);
258	mdelay(33);
259	PSURGE_QUAD_BIC(PSURGE_QUAD_RESET_CTL, procbits);
260	mdelay(33);
261	PSURGE_QUAD_BIS(PSURGE_QUAD_PRIMARY_ARB, procbits);
262	mdelay(33);
263}
264
265static int __init smp_psurge_probe(void)
266{
267	int i, ncpus;
268	struct device_node *dn;
269
270	/* We don't do SMP on the PPC601 -- paulus */
271	if (PVR_VER(mfspr(SPRN_PVR)) == 1)
272		return 1;
273
274	/*
275	 * The powersurge cpu board can be used in the generation
276	 * of powermacs that have a socket for an upgradeable cpu card,
277	 * including the 7500, 8500, 9500, 9600.
278	 * The device tree doesn't tell you if you have 2 cpus because
279	 * OF doesn't know anything about the 2nd processor.
280	 * Instead we look for magic bits in magic registers,
281	 * in the hammerhead memory controller in the case of the
282	 * dual-cpu powersurge board.  -- paulus.
283	 */
284	dn = of_find_node_by_name(NULL, "hammerhead");
285	if (dn == NULL)
286		return 1;
287	of_node_put(dn);
288
289	hhead_base = ioremap(HAMMERHEAD_BASE, 0x800);
290	quad_base = ioremap(PSURGE_QUAD_REG_ADDR, 1024);
291	psurge_sec_intr = hhead_base + HHEAD_SEC_INTR;
292
293	psurge_type = psurge_quad_probe();
294	if (psurge_type != PSURGE_DUAL) {
295		psurge_quad_init();
296		/* All released cards using this HW design have 4 CPUs */
297		ncpus = 4;
298		/* No sure how timebase sync works on those, let's use SW */
299		smp_ops->give_timebase = smp_generic_give_timebase;
300		smp_ops->take_timebase = smp_generic_take_timebase;
301	} else {
302		iounmap(quad_base);
303		if ((in_8(hhead_base + HHEAD_CONFIG) & 0x02) == 0) {
304			/* not a dual-cpu card */
305			iounmap(hhead_base);
306			psurge_type = PSURGE_NONE;
307			return 1;
308		}
309		ncpus = 2;
310	}
311
312	psurge_start = ioremap(PSURGE_START, 4);
313	psurge_pri_intr = ioremap(PSURGE_PRI_INTR, 4);
314
315	/* This is necessary because OF doesn't know about the
316	 * secondary cpu(s), and thus there aren't nodes in the
317	 * device tree for them, and smp_setup_cpu_maps hasn't
318	 * set their bits in cpu_present_map.
319	 */
320	if (ncpus > NR_CPUS)
321		ncpus = NR_CPUS;
322	for (i = 1; i < ncpus ; ++i)
323		set_cpu_present(i, true);
324
325	if (ppc_md.progress) ppc_md.progress("smp_psurge_probe - done", 0x352);
326
327	return ncpus;
328}
329
330static void __init smp_psurge_kick_cpu(int nr)
331{
332	unsigned long start = __pa(__secondary_start_pmac_0) + nr * 8;
333	unsigned long a, flags;
334	int i, j;
335
336	/* Defining this here is evil ... but I prefer hiding that
337	 * crap to avoid giving people ideas that they can do the
338	 * same.
339	 */
340	extern volatile unsigned int cpu_callin_map[NR_CPUS];
341
342	/* may need to flush here if secondary bats aren't setup */
343	for (a = KERNELBASE; a < KERNELBASE + 0x800000; a += 32)
344		asm volatile("dcbf 0,%0" : : "r" (a) : "memory");
345	asm volatile("sync");
346
347	if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu", 0x353);
348
349	/* This is going to freeze the timeebase, we disable interrupts */
350	local_irq_save(flags);
351
352	out_be32(psurge_start, start);
353	mb();
354
355	psurge_set_ipi(nr);
356
357	/*
358	 * We can't use udelay here because the timebase is now frozen.
359	 */
360	for (i = 0; i < 2000; ++i)
361		asm volatile("nop" : : : "memory");
362	psurge_clr_ipi(nr);
363
364	/*
365	 * Also, because the timebase is frozen, we must not return to the
366	 * caller which will try to do udelay's etc... Instead, we wait -here-
367	 * for the CPU to callin.
368	 */
369	for (i = 0; i < 100000 && !cpu_callin_map[nr]; ++i) {
370		for (j = 1; j < 10000; j++)
371			asm volatile("nop" : : : "memory");
372		asm volatile("sync" : : : "memory");
373	}
374	if (!cpu_callin_map[nr])
375		goto stuck;
376
377	/* And we do the TB sync here too for standard dual CPU cards */
378	if (psurge_type == PSURGE_DUAL) {
379		while(!tb_req)
380			barrier();
381		tb_req = 0;
382		mb();
383		timebase = get_tb();
384		mb();
385		while (timebase)
386			barrier();
387		mb();
388	}
389 stuck:
390	/* now interrupt the secondary, restarting both TBs */
391	if (psurge_type == PSURGE_DUAL)
392		psurge_set_ipi(1);
393
394	if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu - done", 0x354);
395}
396
397static struct irqaction psurge_irqaction = {
398	.handler = psurge_primary_intr,
399	.flags = IRQF_DISABLED,
400	.name = "primary IPI",
401};
402
403static void __init smp_psurge_setup_cpu(int cpu_nr)
404{
405	if (cpu_nr != 0)
406		return;
407
408	/* reset the entry point so if we get another intr we won't
409	 * try to startup again */
410	out_be32(psurge_start, 0x100);
411	if (setup_irq(irq_create_mapping(NULL, 30), &psurge_irqaction))
412		printk(KERN_ERR "Couldn't get primary IPI interrupt");
413}
414
415void __init smp_psurge_take_timebase(void)
416{
417	if (psurge_type != PSURGE_DUAL)
418		return;
419
420	tb_req = 1;
421	mb();
422	while (!timebase)
423		barrier();
424	mb();
425	set_tb(timebase >> 32, timebase & 0xffffffff);
426	timebase = 0;
427	mb();
428	set_dec(tb_ticks_per_jiffy/2);
429}
430
431void __init smp_psurge_give_timebase(void)
432{
433	/* Nothing to do here */
434}
435
436/* PowerSurge-style Macs */
437struct smp_ops_t psurge_smp_ops = {
438	.message_pass	= smp_psurge_message_pass,
439	.probe		= smp_psurge_probe,
440	.kick_cpu	= smp_psurge_kick_cpu,
441	.setup_cpu	= smp_psurge_setup_cpu,
442	.give_timebase	= smp_psurge_give_timebase,
443	.take_timebase	= smp_psurge_take_timebase,
444};
445#endif /* CONFIG_PPC32 - actually powersurge support */
446
447/*
448 * Core 99 and later support
449 */
450
451
452static void smp_core99_give_timebase(void)
453{
454	unsigned long flags;
455
456	local_irq_save(flags);
457
458	while(!tb_req)
459		barrier();
460	tb_req = 0;
461	(*pmac_tb_freeze)(1);
462	mb();
463	timebase = get_tb();
464	mb();
465	while (timebase)
466		barrier();
467	mb();
468	(*pmac_tb_freeze)(0);
469	mb();
470
471	local_irq_restore(flags);
472}
473
474
475static void __devinit smp_core99_take_timebase(void)
476{
477	unsigned long flags;
478
479	local_irq_save(flags);
480
481	tb_req = 1;
482	mb();
483	while (!timebase)
484		barrier();
485	mb();
486	set_tb(timebase >> 32, timebase & 0xffffffff);
487	timebase = 0;
488	mb();
489
490	local_irq_restore(flags);
491}
492
493#ifdef CONFIG_PPC64
494/*
495 * G5s enable/disable the timebase via an i2c-connected clock chip.
496 */
497static struct pmac_i2c_bus *pmac_tb_clock_chip_host;
498static u8 pmac_tb_pulsar_addr;
499
500static void smp_core99_cypress_tb_freeze(int freeze)
501{
502	u8 data;
503	int rc;
504
505	/* Strangely, the device-tree says address is 0xd2, but darwin
506	 * accesses 0xd0 ...
507	 */
508	pmac_i2c_setmode(pmac_tb_clock_chip_host,
509			 pmac_i2c_mode_combined);
510	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
511			   0xd0 | pmac_i2c_read,
512			   1, 0x81, &data, 1);
513	if (rc != 0)
514		goto bail;
515
516	data = (data & 0xf3) | (freeze ? 0x00 : 0x0c);
517
518       	pmac_i2c_setmode(pmac_tb_clock_chip_host, pmac_i2c_mode_stdsub);
519	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
520			   0xd0 | pmac_i2c_write,
521			   1, 0x81, &data, 1);
522
523 bail:
524	if (rc != 0) {
525		printk("Cypress Timebase %s rc: %d\n",
526		       freeze ? "freeze" : "unfreeze", rc);
527		panic("Timebase freeze failed !\n");
528	}
529}
530
531
532static void smp_core99_pulsar_tb_freeze(int freeze)
533{
534	u8 data;
535	int rc;
536
537	pmac_i2c_setmode(pmac_tb_clock_chip_host,
538			 pmac_i2c_mode_combined);
539	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
540			   pmac_tb_pulsar_addr | pmac_i2c_read,
541			   1, 0x2e, &data, 1);
542	if (rc != 0)
543		goto bail;
544
545	data = (data & 0x88) | (freeze ? 0x11 : 0x22);
546
547	pmac_i2c_setmode(pmac_tb_clock_chip_host, pmac_i2c_mode_stdsub);
548	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
549			   pmac_tb_pulsar_addr | pmac_i2c_write,
550			   1, 0x2e, &data, 1);
551 bail:
552	if (rc != 0) {
553		printk(KERN_ERR "Pulsar Timebase %s rc: %d\n",
554		       freeze ? "freeze" : "unfreeze", rc);
555		panic("Timebase freeze failed !\n");
556	}
557}
558
559static void __init smp_core99_setup_i2c_hwsync(int ncpus)
560{
561	struct device_node *cc = NULL;	
562	struct device_node *p;
563	const char *name = NULL;
564	const u32 *reg;
565	int ok;
566
567	/* Look for the clock chip */
568	while ((cc = of_find_node_by_name(cc, "i2c-hwclock")) != NULL) {
569		p = of_get_parent(cc);
570		ok = p && of_device_is_compatible(p, "uni-n-i2c");
571		of_node_put(p);
572		if (!ok)
573			continue;
574
575		pmac_tb_clock_chip_host = pmac_i2c_find_bus(cc);
576		if (pmac_tb_clock_chip_host == NULL)
577			continue;
578		reg = of_get_property(cc, "reg", NULL);
579		if (reg == NULL)
580			continue;
581		switch (*reg) {
582		case 0xd2:
583			if (of_device_is_compatible(cc,"pulsar-legacy-slewing")) {
584				pmac_tb_freeze = smp_core99_pulsar_tb_freeze;
585				pmac_tb_pulsar_addr = 0xd2;
586				name = "Pulsar";
587			} else if (of_device_is_compatible(cc, "cy28508")) {
588				pmac_tb_freeze = smp_core99_cypress_tb_freeze;
589				name = "Cypress";
590			}
591			break;
592		case 0xd4:
593			pmac_tb_freeze = smp_core99_pulsar_tb_freeze;
594			pmac_tb_pulsar_addr = 0xd4;
595			name = "Pulsar";
596			break;
597		}
598		if (pmac_tb_freeze != NULL)
599			break;
600	}
601	if (pmac_tb_freeze != NULL) {
602		/* Open i2c bus for synchronous access */
603		if (pmac_i2c_open(pmac_tb_clock_chip_host, 1)) {
604			printk(KERN_ERR "Failed top open i2c bus for clock"
605			       " sync, fallback to software sync !\n");
606			goto no_i2c_sync;
607		}
608		printk(KERN_INFO "Processor timebase sync using %s i2c clock\n",
609		       name);
610		return;
611	}
612 no_i2c_sync:
613	pmac_tb_freeze = NULL;
614	pmac_tb_clock_chip_host = NULL;
615}
616
617
618
619/*
620 * Newer G5s uses a platform function
621 */
622
623static void smp_core99_pfunc_tb_freeze(int freeze)
624{
625	struct device_node *cpus;
626	struct pmf_args args;
627
628	cpus = of_find_node_by_path("/cpus");
629	BUG_ON(cpus == NULL);
630	args.count = 1;
631	args.u[0].v = !freeze;
632	pmf_call_function(cpus, "cpu-timebase", &args);
633	of_node_put(cpus);
634}
635
636#else /* CONFIG_PPC64 */
637
638/*
639 * SMP G4 use a GPIO to enable/disable the timebase.
640 */
641
642static unsigned int core99_tb_gpio;	/* Timebase freeze GPIO */
643
644static void smp_core99_gpio_tb_freeze(int freeze)
645{
646	if (freeze)
647		pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 4);
648	else
649		pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 0);
650	pmac_call_feature(PMAC_FTR_READ_GPIO, NULL, core99_tb_gpio, 0);
651}
652
653
654#endif /* !CONFIG_PPC64 */
655
656/* L2 and L3 cache settings to pass from CPU0 to CPU1 on G4 cpus */
657volatile static long int core99_l2_cache;
658volatile static long int core99_l3_cache;
659
660static void __devinit core99_init_caches(int cpu)
661{
662#ifndef CONFIG_PPC64
663	if (!cpu_has_feature(CPU_FTR_L2CR))
664		return;
665
666	if (cpu == 0) {
667		core99_l2_cache = _get_L2CR();
668		printk("CPU0: L2CR is %lx\n", core99_l2_cache);
669	} else {
670		printk("CPU%d: L2CR was %lx\n", cpu, _get_L2CR());
671		_set_L2CR(0);
672		_set_L2CR(core99_l2_cache);
673		printk("CPU%d: L2CR set to %lx\n", cpu, core99_l2_cache);
674	}
675
676	if (!cpu_has_feature(CPU_FTR_L3CR))
677		return;
678
679	if (cpu == 0){
680		core99_l3_cache = _get_L3CR();
681		printk("CPU0: L3CR is %lx\n", core99_l3_cache);
682	} else {
683		printk("CPU%d: L3CR was %lx\n", cpu, _get_L3CR());
684		_set_L3CR(0);
685		_set_L3CR(core99_l3_cache);
686		printk("CPU%d: L3CR set to %lx\n", cpu, core99_l3_cache);
687	}
688#endif /* !CONFIG_PPC64 */
689}
690
691static void __init smp_core99_setup(int ncpus)
692{
693#ifdef CONFIG_PPC64
694
695	/* i2c based HW sync on some G5s */
696	if (of_machine_is_compatible("PowerMac7,2") ||
697	    of_machine_is_compatible("PowerMac7,3") ||
698	    of_machine_is_compatible("RackMac3,1"))
699		smp_core99_setup_i2c_hwsync(ncpus);
700
701	/* pfunc based HW sync on recent G5s */
702	if (pmac_tb_freeze == NULL) {
703		struct device_node *cpus =
704			of_find_node_by_path("/cpus");
705		if (cpus &&
706		    of_get_property(cpus, "platform-cpu-timebase", NULL)) {
707			pmac_tb_freeze = smp_core99_pfunc_tb_freeze;
708			printk(KERN_INFO "Processor timebase sync using"
709			       " platform function\n");
710		}
711	}
712
713#else /* CONFIG_PPC64 */
714
715	/* GPIO based HW sync on ppc32 Core99 */
716	if (pmac_tb_freeze == NULL && !of_machine_is_compatible("MacRISC4")) {
717		struct device_node *cpu;
718		const u32 *tbprop = NULL;
719
720		core99_tb_gpio = KL_GPIO_TB_ENABLE;	/* default value */
721		cpu = of_find_node_by_type(NULL, "cpu");
722		if (cpu != NULL) {
723			tbprop = of_get_property(cpu, "timebase-enable", NULL);
724			if (tbprop)
725				core99_tb_gpio = *tbprop;
726			of_node_put(cpu);
727		}
728		pmac_tb_freeze = smp_core99_gpio_tb_freeze;
729		printk(KERN_INFO "Processor timebase sync using"
730		       " GPIO 0x%02x\n", core99_tb_gpio);
731	}
732
733#endif /* CONFIG_PPC64 */
734
735	/* No timebase sync, fallback to software */
736	if (pmac_tb_freeze == NULL) {
737		smp_ops->give_timebase = smp_generic_give_timebase;
738		smp_ops->take_timebase = smp_generic_take_timebase;
739		printk(KERN_INFO "Processor timebase sync using software\n");
740	}
741
742#ifndef CONFIG_PPC64
743	{
744		int i;
745
746		/* XXX should get this from reg properties */
747		for (i = 1; i < ncpus; ++i)
748			set_hard_smp_processor_id(i, i);
749	}
750#endif
751
752	/* 32 bits SMP can't NAP */
753	if (!of_machine_is_compatible("MacRISC4"))
754		powersave_nap = 0;
755}
756
757static int __init smp_core99_probe(void)
758{
759	struct device_node *cpus;
760	int ncpus = 0;
761
762	if (ppc_md.progress) ppc_md.progress("smp_core99_probe", 0x345);
763
764	/* Count CPUs in the device-tree */
765       	for (cpus = NULL; (cpus = of_find_node_by_type(cpus, "cpu")) != NULL;)
766	       	++ncpus;
767
768	printk(KERN_INFO "PowerMac SMP probe found %d cpus\n", ncpus);
769
770	/* Nothing more to do if less than 2 of them */
771	if (ncpus <= 1)
772		return 1;
773
774	/* We need to perform some early initialisations before we can start
775	 * setting up SMP as we are running before initcalls
776	 */
777	pmac_pfunc_base_install();
778	pmac_i2c_init();
779
780	/* Setup various bits like timebase sync method, ability to nap, ... */
781	smp_core99_setup(ncpus);
782
783	/* Install IPIs */
784	mpic_request_ipis();
785
786	/* Collect l2cr and l3cr values from CPU 0 */
787	core99_init_caches(0);
788
789	return ncpus;
790}
791
792static void __devinit smp_core99_kick_cpu(int nr)
793{
794	unsigned int save_vector;
795	unsigned long target, flags;
796	unsigned int *vector = (unsigned int *)(PAGE_OFFSET+0x100);
797
798	if (nr < 0 || nr > 3)
799		return;
800
801	if (ppc_md.progress)
802		ppc_md.progress("smp_core99_kick_cpu", 0x346);
803
804	local_irq_save(flags);
805
806	/* Save reset vector */
807	save_vector = *vector;
808
809	/* Setup fake reset vector that does
810	 *   b __secondary_start_pmac_0 + nr*8
811	 */
812	target = (unsigned long) __secondary_start_pmac_0 + nr * 8;
813	patch_branch(vector, target, BRANCH_SET_LINK);
814
815	/* Put some life in our friend */
816	pmac_call_feature(PMAC_FTR_RESET_CPU, NULL, nr, 0);
817
818	/* FIXME: We wait a bit for the CPU to take the exception, I should
819	 * instead wait for the entry code to set something for me. Well,
820	 * ideally, all that crap will be done in prom.c and the CPU left
821	 * in a RAM-based wait loop like CHRP.
822	 */
823	mdelay(1);
824
825	/* Restore our exception vector */
826	*vector = save_vector;
827	flush_icache_range((unsigned long) vector, (unsigned long) vector + 4);
828
829	local_irq_restore(flags);
830	if (ppc_md.progress) ppc_md.progress("smp_core99_kick_cpu done", 0x347);
831}
832
833static void __devinit smp_core99_setup_cpu(int cpu_nr)
834{
835	/* Setup L2/L3 */
836	if (cpu_nr != 0)
837		core99_init_caches(cpu_nr);
838
839	/* Setup openpic */
840	mpic_setup_this_cpu();
841
842	if (cpu_nr == 0) {
843#ifdef CONFIG_PPC64
844		extern void g5_phy_disable_cpu1(void);
845
846		/* Close i2c bus if it was used for tb sync */
847		if (pmac_tb_clock_chip_host) {
848			pmac_i2c_close(pmac_tb_clock_chip_host);
849			pmac_tb_clock_chip_host	= NULL;
850		}
851
852		/* If we didn't start the second CPU, we must take
853		 * it off the bus
854		 */
855		if (of_machine_is_compatible("MacRISC4") &&
856		    num_online_cpus() < 2)		
857			g5_phy_disable_cpu1();
858#endif /* CONFIG_PPC64 */
859
860		if (ppc_md.progress)
861			ppc_md.progress("core99_setup_cpu 0 done", 0x349);
862	}
863}
864
865
866#if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PPC32)
867
868int smp_core99_cpu_disable(void)
869{
870	set_cpu_online(smp_processor_id(), false);
871
872	/* XXX reset cpu affinity here */
873	mpic_cpu_set_priority(0xf);
874	asm volatile("mtdec %0" : : "r" (0x7fffffff));
875	mb();
876	udelay(20);
877	asm volatile("mtdec %0" : : "r" (0x7fffffff));
878	return 0;
879}
880
881extern void low_cpu_die(void) __attribute__((noreturn)); /* in sleep.S */
882static int cpu_dead[NR_CPUS];
883
884void cpu_die(void)
885{
886	local_irq_disable();
887	cpu_dead[smp_processor_id()] = 1;
888	mb();
889	low_cpu_die();
890}
891
892void smp_core99_cpu_die(unsigned int cpu)
893{
894	int timeout;
895
896	timeout = 1000;
897	while (!cpu_dead[cpu]) {
898		if (--timeout == 0) {
899			printk("CPU %u refused to die!\n", cpu);
900			break;
901		}
902		msleep(1);
903	}
904	cpu_dead[cpu] = 0;
905}
906
907#endif /* CONFIG_HOTPLUG_CPU && CONFIG_PP32 */
908
909/* Core99 Macs (dual G4s and G5s) */
910struct smp_ops_t core99_smp_ops = {
911	.message_pass	= smp_mpic_message_pass,
912	.probe		= smp_core99_probe,
913	.kick_cpu	= smp_core99_kick_cpu,
914	.setup_cpu	= smp_core99_setup_cpu,
915	.give_timebase	= smp_core99_give_timebase,
916	.take_timebase	= smp_core99_take_timebase,
917#if defined(CONFIG_HOTPLUG_CPU)
918# if defined(CONFIG_PPC32)
919	.cpu_disable	= smp_core99_cpu_disable,
920	.cpu_die	= smp_core99_cpu_die,
921# endif
922# if defined(CONFIG_PPC64)
923	.cpu_disable	= generic_cpu_disable,
924	.cpu_die	= generic_cpu_die,
925	/* intentionally do *NOT* assign cpu_enable,
926	 * the generic code will use kick_cpu then! */
927# endif
928#endif
929};
930
931void __init pmac_setup_smp(void)
932{
933	struct device_node *np;
934
935	/* Check for Core99 */
936	np = of_find_node_by_name(NULL, "uni-n");
937	if (!np)
938		np = of_find_node_by_name(NULL, "u3");
939	if (!np)
940		np = of_find_node_by_name(NULL, "u4");
941	if (np) {
942		of_node_put(np);
943		smp_ops = &core99_smp_ops;
944	}
945#ifdef CONFIG_PPC32
946	else {
947		/* We have to set bits in cpu_possible_map here since the
948		 * secondary CPU(s) aren't in the device tree. Various
949		 * things won't be initialized for CPUs not in the possible
950		 * map, so we really need to fix it up here.
951		 */
952		int cpu;
953
954		for (cpu = 1; cpu < 4 && cpu < NR_CPUS; ++cpu)
955			set_cpu_possible(cpu, true);
956		smp_ops = &psurge_smp_ops;
957	}
958#endif /* CONFIG_PPC32 */
959}
960