Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/linville/wirel...
[pandora-kernel.git] / arch / powerpc / platforms / powermac / smp.c
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 #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
66 extern void __secondary_start_pmac_0(void);
67 extern int pmac_pfunc_base_install(void);
68
69 static void (*pmac_tb_freeze)(int freeze);
70 static u64 timebase;
71 static int tb_req;
72
73 #ifdef CONFIG_PPC32
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 */
111 static volatile u8 __iomem *hhead_base;
112 static volatile u8 __iomem *quad_base;
113 static volatile u32 __iomem *psurge_pri_intr;
114 static volatile u8 __iomem *psurge_sec_intr;
115 static 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 */
125 static int psurge_type = PSURGE_NONE;
126
127 /*
128  * Set and clear IPIs for powersurge.
129  */
130 static inline void psurge_set_ipi(int cpu)
131 {
132         if (psurge_type == PSURGE_NONE)
133                 return;
134         if (cpu == 0)
135                 in_be32(psurge_pri_intr);
136         else if (psurge_type == PSURGE_DUAL)
137                 out_8(psurge_sec_intr, 0);
138         else
139                 PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_SET, 1 << cpu);
140 }
141
142 static inline void psurge_clr_ipi(int cpu)
143 {
144         if (cpu > 0) {
145                 switch(psurge_type) {
146                 case PSURGE_DUAL:
147                         out_8(psurge_sec_intr, ~0);
148                 case PSURGE_NONE:
149                         break;
150                 default:
151                         PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_CLR, 1 << cpu);
152                 }
153         }
154 }
155
156 /*
157  * On powersurge (old SMP powermac architecture) we don't have
158  * separate IPIs for separate messages like openpic does.  Instead
159  * we have a bitmap for each processor, where a 1 bit means that
160  * the corresponding message is pending for that processor.
161  * Ideally each cpu's entry would be in a different cache line.
162  *  -- paulus.
163  */
164 static unsigned long psurge_smp_message[NR_CPUS];
165
166 void psurge_smp_message_recv(void)
167 {
168         int cpu = smp_processor_id();
169         int msg;
170
171         /* clear interrupt */
172         psurge_clr_ipi(cpu);
173
174         if (num_online_cpus() < 2)
175                 return;
176
177         /* make sure there is a message there */
178         for (msg = 0; msg < 4; msg++)
179                 if (test_and_clear_bit(msg, &psurge_smp_message[cpu]))
180                         smp_message_recv(msg);
181 }
182
183 irqreturn_t psurge_primary_intr(int irq, void *d)
184 {
185         psurge_smp_message_recv();
186         return IRQ_HANDLED;
187 }
188
189 static void smp_psurge_message_pass(int target, int msg)
190 {
191         int i;
192
193         if (num_online_cpus() < 2)
194                 return;
195
196         for_each_online_cpu(i) {
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         struct device_node *dn;
271
272         /* We don't do SMP on the PPC601 -- paulus */
273         if (PVR_VER(mfspr(SPRN_PVR)) == 1)
274                 return 1;
275
276         /*
277          * The powersurge cpu board can be used in the generation
278          * of powermacs that have a socket for an upgradeable cpu card,
279          * including the 7500, 8500, 9500, 9600.
280          * The device tree doesn't tell you if you have 2 cpus because
281          * OF doesn't know anything about the 2nd processor.
282          * Instead we look for magic bits in magic registers,
283          * in the hammerhead memory controller in the case of the
284          * dual-cpu powersurge board.  -- paulus.
285          */
286         dn = of_find_node_by_name(NULL, "hammerhead");
287         if (dn == NULL)
288                 return 1;
289         of_node_put(dn);
290
291         hhead_base = ioremap(HAMMERHEAD_BASE, 0x800);
292         quad_base = ioremap(PSURGE_QUAD_REG_ADDR, 1024);
293         psurge_sec_intr = hhead_base + HHEAD_SEC_INTR;
294
295         psurge_type = psurge_quad_probe();
296         if (psurge_type != PSURGE_DUAL) {
297                 psurge_quad_init();
298                 /* All released cards using this HW design have 4 CPUs */
299                 ncpus = 4;
300                 /* No sure how timebase sync works on those, let's use SW */
301                 smp_ops->give_timebase = smp_generic_give_timebase;
302                 smp_ops->take_timebase = smp_generic_take_timebase;
303         } else {
304                 iounmap(quad_base);
305                 if ((in_8(hhead_base + HHEAD_CONFIG) & 0x02) == 0) {
306                         /* not a dual-cpu card */
307                         iounmap(hhead_base);
308                         psurge_type = PSURGE_NONE;
309                         return 1;
310                 }
311                 ncpus = 2;
312         }
313
314         psurge_start = ioremap(PSURGE_START, 4);
315         psurge_pri_intr = ioremap(PSURGE_PRI_INTR, 4);
316
317         /* This is necessary because OF doesn't know about the
318          * secondary cpu(s), and thus there aren't nodes in the
319          * device tree for them, and smp_setup_cpu_maps hasn't
320          * set their bits in cpu_present_mask.
321          */
322         if (ncpus > NR_CPUS)
323                 ncpus = NR_CPUS;
324         for (i = 1; i < ncpus ; ++i)
325                 set_cpu_present(i, true);
326
327         if (ppc_md.progress) ppc_md.progress("smp_psurge_probe - done", 0x352);
328
329         return ncpus;
330 }
331
332 static void __init smp_psurge_kick_cpu(int nr)
333 {
334         unsigned long start = __pa(__secondary_start_pmac_0) + nr * 8;
335         unsigned long a, flags;
336         int i, j;
337
338         /* Defining this here is evil ... but I prefer hiding that
339          * crap to avoid giving people ideas that they can do the
340          * same.
341          */
342         extern volatile unsigned int cpu_callin_map[NR_CPUS];
343
344         /* may need to flush here if secondary bats aren't setup */
345         for (a = KERNELBASE; a < KERNELBASE + 0x800000; a += 32)
346                 asm volatile("dcbf 0,%0" : : "r" (a) : "memory");
347         asm volatile("sync");
348
349         if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu", 0x353);
350
351         /* This is going to freeze the timeebase, we disable interrupts */
352         local_irq_save(flags);
353
354         out_be32(psurge_start, start);
355         mb();
356
357         psurge_set_ipi(nr);
358
359         /*
360          * We can't use udelay here because the timebase is now frozen.
361          */
362         for (i = 0; i < 2000; ++i)
363                 asm volatile("nop" : : : "memory");
364         psurge_clr_ipi(nr);
365
366         /*
367          * Also, because the timebase is frozen, we must not return to the
368          * caller which will try to do udelay's etc... Instead, we wait -here-
369          * for the CPU to callin.
370          */
371         for (i = 0; i < 100000 && !cpu_callin_map[nr]; ++i) {
372                 for (j = 1; j < 10000; j++)
373                         asm volatile("nop" : : : "memory");
374                 asm volatile("sync" : : : "memory");
375         }
376         if (!cpu_callin_map[nr])
377                 goto stuck;
378
379         /* And we do the TB sync here too for standard dual CPU cards */
380         if (psurge_type == PSURGE_DUAL) {
381                 while(!tb_req)
382                         barrier();
383                 tb_req = 0;
384                 mb();
385                 timebase = get_tb();
386                 mb();
387                 while (timebase)
388                         barrier();
389                 mb();
390         }
391  stuck:
392         /* now interrupt the secondary, restarting both TBs */
393         if (psurge_type == PSURGE_DUAL)
394                 psurge_set_ipi(1);
395
396         if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu - done", 0x354);
397 }
398
399 static struct irqaction psurge_irqaction = {
400         .handler = psurge_primary_intr,
401         .flags = IRQF_DISABLED,
402         .name = "primary IPI",
403 };
404
405 static void __init smp_psurge_setup_cpu(int cpu_nr)
406 {
407         if (cpu_nr != 0)
408                 return;
409
410         /* reset the entry point so if we get another intr we won't
411          * try to startup again */
412         out_be32(psurge_start, 0x100);
413         if (setup_irq(irq_create_mapping(NULL, 30), &psurge_irqaction))
414                 printk(KERN_ERR "Couldn't get primary IPI interrupt");
415 }
416
417 void __init smp_psurge_take_timebase(void)
418 {
419         if (psurge_type != PSURGE_DUAL)
420                 return;
421
422         tb_req = 1;
423         mb();
424         while (!timebase)
425                 barrier();
426         mb();
427         set_tb(timebase >> 32, timebase & 0xffffffff);
428         timebase = 0;
429         mb();
430         set_dec(tb_ticks_per_jiffy/2);
431 }
432
433 void __init smp_psurge_give_timebase(void)
434 {
435         /* Nothing to do here */
436 }
437
438 /* PowerSurge-style Macs */
439 struct smp_ops_t psurge_smp_ops = {
440         .message_pass   = smp_psurge_message_pass,
441         .probe          = smp_psurge_probe,
442         .kick_cpu       = smp_psurge_kick_cpu,
443         .setup_cpu      = smp_psurge_setup_cpu,
444         .give_timebase  = smp_psurge_give_timebase,
445         .take_timebase  = smp_psurge_take_timebase,
446 };
447 #endif /* CONFIG_PPC32 - actually powersurge support */
448
449 /*
450  * Core 99 and later support
451  */
452
453
454 static void smp_core99_give_timebase(void)
455 {
456         unsigned long flags;
457
458         local_irq_save(flags);
459
460         while(!tb_req)
461                 barrier();
462         tb_req = 0;
463         (*pmac_tb_freeze)(1);
464         mb();
465         timebase = get_tb();
466         mb();
467         while (timebase)
468                 barrier();
469         mb();
470         (*pmac_tb_freeze)(0);
471         mb();
472
473         local_irq_restore(flags);
474 }
475
476
477 static void __devinit smp_core99_take_timebase(void)
478 {
479         unsigned long flags;
480
481         local_irq_save(flags);
482
483         tb_req = 1;
484         mb();
485         while (!timebase)
486                 barrier();
487         mb();
488         set_tb(timebase >> 32, timebase & 0xffffffff);
489         timebase = 0;
490         mb();
491
492         local_irq_restore(flags);
493 }
494
495 #ifdef CONFIG_PPC64
496 /*
497  * G5s enable/disable the timebase via an i2c-connected clock chip.
498  */
499 static struct pmac_i2c_bus *pmac_tb_clock_chip_host;
500 static u8 pmac_tb_pulsar_addr;
501
502 static void smp_core99_cypress_tb_freeze(int freeze)
503 {
504         u8 data;
505         int rc;
506
507         /* Strangely, the device-tree says address is 0xd2, but darwin
508          * accesses 0xd0 ...
509          */
510         pmac_i2c_setmode(pmac_tb_clock_chip_host,
511                          pmac_i2c_mode_combined);
512         rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
513                            0xd0 | pmac_i2c_read,
514                            1, 0x81, &data, 1);
515         if (rc != 0)
516                 goto bail;
517
518         data = (data & 0xf3) | (freeze ? 0x00 : 0x0c);
519
520         pmac_i2c_setmode(pmac_tb_clock_chip_host, pmac_i2c_mode_stdsub);
521         rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
522                            0xd0 | pmac_i2c_write,
523                            1, 0x81, &data, 1);
524
525  bail:
526         if (rc != 0) {
527                 printk("Cypress Timebase %s rc: %d\n",
528                        freeze ? "freeze" : "unfreeze", rc);
529                 panic("Timebase freeze failed !\n");
530         }
531 }
532
533
534 static void smp_core99_pulsar_tb_freeze(int freeze)
535 {
536         u8 data;
537         int rc;
538
539         pmac_i2c_setmode(pmac_tb_clock_chip_host,
540                          pmac_i2c_mode_combined);
541         rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
542                            pmac_tb_pulsar_addr | pmac_i2c_read,
543                            1, 0x2e, &data, 1);
544         if (rc != 0)
545                 goto bail;
546
547         data = (data & 0x88) | (freeze ? 0x11 : 0x22);
548
549         pmac_i2c_setmode(pmac_tb_clock_chip_host, pmac_i2c_mode_stdsub);
550         rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
551                            pmac_tb_pulsar_addr | pmac_i2c_write,
552                            1, 0x2e, &data, 1);
553  bail:
554         if (rc != 0) {
555                 printk(KERN_ERR "Pulsar Timebase %s rc: %d\n",
556                        freeze ? "freeze" : "unfreeze", rc);
557                 panic("Timebase freeze failed !\n");
558         }
559 }
560
561 static void __init smp_core99_setup_i2c_hwsync(int ncpus)
562 {
563         struct device_node *cc = NULL;  
564         struct device_node *p;
565         const char *name = NULL;
566         const u32 *reg;
567         int ok;
568
569         /* Look for the clock chip */
570         while ((cc = of_find_node_by_name(cc, "i2c-hwclock")) != NULL) {
571                 p = of_get_parent(cc);
572                 ok = p && of_device_is_compatible(p, "uni-n-i2c");
573                 of_node_put(p);
574                 if (!ok)
575                         continue;
576
577                 pmac_tb_clock_chip_host = pmac_i2c_find_bus(cc);
578                 if (pmac_tb_clock_chip_host == NULL)
579                         continue;
580                 reg = of_get_property(cc, "reg", NULL);
581                 if (reg == NULL)
582                         continue;
583                 switch (*reg) {
584                 case 0xd2:
585                         if (of_device_is_compatible(cc,"pulsar-legacy-slewing")) {
586                                 pmac_tb_freeze = smp_core99_pulsar_tb_freeze;
587                                 pmac_tb_pulsar_addr = 0xd2;
588                                 name = "Pulsar";
589                         } else if (of_device_is_compatible(cc, "cy28508")) {
590                                 pmac_tb_freeze = smp_core99_cypress_tb_freeze;
591                                 name = "Cypress";
592                         }
593                         break;
594                 case 0xd4:
595                         pmac_tb_freeze = smp_core99_pulsar_tb_freeze;
596                         pmac_tb_pulsar_addr = 0xd4;
597                         name = "Pulsar";
598                         break;
599                 }
600                 if (pmac_tb_freeze != NULL)
601                         break;
602         }
603         if (pmac_tb_freeze != NULL) {
604                 /* Open i2c bus for synchronous access */
605                 if (pmac_i2c_open(pmac_tb_clock_chip_host, 1)) {
606                         printk(KERN_ERR "Failed top open i2c bus for clock"
607                                " sync, fallback to software sync !\n");
608                         goto no_i2c_sync;
609                 }
610                 printk(KERN_INFO "Processor timebase sync using %s i2c clock\n",
611                        name);
612                 return;
613         }
614  no_i2c_sync:
615         pmac_tb_freeze = NULL;
616         pmac_tb_clock_chip_host = NULL;
617 }
618
619
620
621 /*
622  * Newer G5s uses a platform function
623  */
624
625 static void smp_core99_pfunc_tb_freeze(int freeze)
626 {
627         struct device_node *cpus;
628         struct pmf_args args;
629
630         cpus = of_find_node_by_path("/cpus");
631         BUG_ON(cpus == NULL);
632         args.count = 1;
633         args.u[0].v = !freeze;
634         pmf_call_function(cpus, "cpu-timebase", &args);
635         of_node_put(cpus);
636 }
637
638 #else /* CONFIG_PPC64 */
639
640 /*
641  * SMP G4 use a GPIO to enable/disable the timebase.
642  */
643
644 static unsigned int core99_tb_gpio;     /* Timebase freeze GPIO */
645
646 static void smp_core99_gpio_tb_freeze(int freeze)
647 {
648         if (freeze)
649                 pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 4);
650         else
651                 pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 0);
652         pmac_call_feature(PMAC_FTR_READ_GPIO, NULL, core99_tb_gpio, 0);
653 }
654
655
656 #endif /* !CONFIG_PPC64 */
657
658 /* L2 and L3 cache settings to pass from CPU0 to CPU1 on G4 cpus */
659 volatile static long int core99_l2_cache;
660 volatile static long int core99_l3_cache;
661
662 static void __devinit core99_init_caches(int cpu)
663 {
664 #ifndef CONFIG_PPC64
665         if (!cpu_has_feature(CPU_FTR_L2CR))
666                 return;
667
668         if (cpu == 0) {
669                 core99_l2_cache = _get_L2CR();
670                 printk("CPU0: L2CR is %lx\n", core99_l2_cache);
671         } else {
672                 printk("CPU%d: L2CR was %lx\n", cpu, _get_L2CR());
673                 _set_L2CR(0);
674                 _set_L2CR(core99_l2_cache);
675                 printk("CPU%d: L2CR set to %lx\n", cpu, core99_l2_cache);
676         }
677
678         if (!cpu_has_feature(CPU_FTR_L3CR))
679                 return;
680
681         if (cpu == 0){
682                 core99_l3_cache = _get_L3CR();
683                 printk("CPU0: L3CR is %lx\n", core99_l3_cache);
684         } else {
685                 printk("CPU%d: L3CR was %lx\n", cpu, _get_L3CR());
686                 _set_L3CR(0);
687                 _set_L3CR(core99_l3_cache);
688                 printk("CPU%d: L3CR set to %lx\n", cpu, core99_l3_cache);
689         }
690 #endif /* !CONFIG_PPC64 */
691 }
692
693 static void __init smp_core99_setup(int ncpus)
694 {
695 #ifdef CONFIG_PPC64
696
697         /* i2c based HW sync on some G5s */
698         if (of_machine_is_compatible("PowerMac7,2") ||
699             of_machine_is_compatible("PowerMac7,3") ||
700             of_machine_is_compatible("RackMac3,1"))
701                 smp_core99_setup_i2c_hwsync(ncpus);
702
703         /* pfunc based HW sync on recent G5s */
704         if (pmac_tb_freeze == NULL) {
705                 struct device_node *cpus =
706                         of_find_node_by_path("/cpus");
707                 if (cpus &&
708                     of_get_property(cpus, "platform-cpu-timebase", NULL)) {
709                         pmac_tb_freeze = smp_core99_pfunc_tb_freeze;
710                         printk(KERN_INFO "Processor timebase sync using"
711                                " platform function\n");
712                 }
713         }
714
715 #else /* CONFIG_PPC64 */
716
717         /* GPIO based HW sync on ppc32 Core99 */
718         if (pmac_tb_freeze == NULL && !of_machine_is_compatible("MacRISC4")) {
719                 struct device_node *cpu;
720                 const u32 *tbprop = NULL;
721
722                 core99_tb_gpio = KL_GPIO_TB_ENABLE;     /* default value */
723                 cpu = of_find_node_by_type(NULL, "cpu");
724                 if (cpu != NULL) {
725                         tbprop = of_get_property(cpu, "timebase-enable", NULL);
726                         if (tbprop)
727                                 core99_tb_gpio = *tbprop;
728                         of_node_put(cpu);
729                 }
730                 pmac_tb_freeze = smp_core99_gpio_tb_freeze;
731                 printk(KERN_INFO "Processor timebase sync using"
732                        " GPIO 0x%02x\n", core99_tb_gpio);
733         }
734
735 #endif /* CONFIG_PPC64 */
736
737         /* No timebase sync, fallback to software */
738         if (pmac_tb_freeze == NULL) {
739                 smp_ops->give_timebase = smp_generic_give_timebase;
740                 smp_ops->take_timebase = smp_generic_take_timebase;
741                 printk(KERN_INFO "Processor timebase sync using software\n");
742         }
743
744 #ifndef CONFIG_PPC64
745         {
746                 int i;
747
748                 /* XXX should get this from reg properties */
749                 for (i = 1; i < ncpus; ++i)
750                         set_hard_smp_processor_id(i, i);
751         }
752 #endif
753
754         /* 32 bits SMP can't NAP */
755         if (!of_machine_is_compatible("MacRISC4"))
756                 powersave_nap = 0;
757 }
758
759 static int __init smp_core99_probe(void)
760 {
761         struct device_node *cpus;
762         int ncpus = 0;
763
764         if (ppc_md.progress) ppc_md.progress("smp_core99_probe", 0x345);
765
766         /* Count CPUs in the device-tree */
767         for (cpus = NULL; (cpus = of_find_node_by_type(cpus, "cpu")) != NULL;)
768                 ++ncpus;
769
770         printk(KERN_INFO "PowerMac SMP probe found %d cpus\n", ncpus);
771
772         /* Nothing more to do if less than 2 of them */
773         if (ncpus <= 1)
774                 return 1;
775
776         /* We need to perform some early initialisations before we can start
777          * setting up SMP as we are running before initcalls
778          */
779         pmac_pfunc_base_install();
780         pmac_i2c_init();
781
782         /* Setup various bits like timebase sync method, ability to nap, ... */
783         smp_core99_setup(ncpus);
784
785         /* Install IPIs */
786         mpic_request_ipis();
787
788         /* Collect l2cr and l3cr values from CPU 0 */
789         core99_init_caches(0);
790
791         return ncpus;
792 }
793
794 static void __devinit smp_core99_kick_cpu(int nr)
795 {
796         unsigned int save_vector;
797         unsigned long target, flags;
798         unsigned int *vector = (unsigned int *)(PAGE_OFFSET+0x100);
799
800         if (nr < 0 || nr > 3)
801                 return;
802
803         if (ppc_md.progress)
804                 ppc_md.progress("smp_core99_kick_cpu", 0x346);
805
806         local_irq_save(flags);
807
808         /* Save reset vector */
809         save_vector = *vector;
810
811         /* Setup fake reset vector that does
812          *   b __secondary_start_pmac_0 + nr*8
813          */
814         target = (unsigned long) __secondary_start_pmac_0 + nr * 8;
815         patch_branch(vector, target, BRANCH_SET_LINK);
816
817         /* Put some life in our friend */
818         pmac_call_feature(PMAC_FTR_RESET_CPU, NULL, nr, 0);
819
820         /* FIXME: We wait a bit for the CPU to take the exception, I should
821          * instead wait for the entry code to set something for me. Well,
822          * ideally, all that crap will be done in prom.c and the CPU left
823          * in a RAM-based wait loop like CHRP.
824          */
825         mdelay(1);
826
827         /* Restore our exception vector */
828         *vector = save_vector;
829         flush_icache_range((unsigned long) vector, (unsigned long) vector + 4);
830
831         local_irq_restore(flags);
832         if (ppc_md.progress) ppc_md.progress("smp_core99_kick_cpu done", 0x347);
833 }
834
835 static void __devinit smp_core99_setup_cpu(int cpu_nr)
836 {
837         /* Setup L2/L3 */
838         if (cpu_nr != 0)
839                 core99_init_caches(cpu_nr);
840
841         /* Setup openpic */
842         mpic_setup_this_cpu();
843
844         if (cpu_nr == 0) {
845 #ifdef CONFIG_PPC64
846                 extern void g5_phy_disable_cpu1(void);
847
848                 /* Close i2c bus if it was used for tb sync */
849                 if (pmac_tb_clock_chip_host) {
850                         pmac_i2c_close(pmac_tb_clock_chip_host);
851                         pmac_tb_clock_chip_host = NULL;
852                 }
853
854                 /* If we didn't start the second CPU, we must take
855                  * it off the bus
856                  */
857                 if (of_machine_is_compatible("MacRISC4") &&
858                     num_online_cpus() < 2)              
859                         g5_phy_disable_cpu1();
860 #endif /* CONFIG_PPC64 */
861
862                 if (ppc_md.progress)
863                         ppc_md.progress("core99_setup_cpu 0 done", 0x349);
864         }
865 }
866
867
868 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PPC32)
869
870 int smp_core99_cpu_disable(void)
871 {
872         set_cpu_online(smp_processor_id(), false);
873
874         /* XXX reset cpu affinity here */
875         mpic_cpu_set_priority(0xf);
876         asm volatile("mtdec %0" : : "r" (0x7fffffff));
877         mb();
878         udelay(20);
879         asm volatile("mtdec %0" : : "r" (0x7fffffff));
880         return 0;
881 }
882
883 static int cpu_dead[NR_CPUS];
884
885 void pmac32_cpu_die(void)
886 {
887         local_irq_disable();
888         cpu_dead[smp_processor_id()] = 1;
889         mb();
890         low_cpu_die();
891 }
892
893 void smp_core99_cpu_die(unsigned int cpu)
894 {
895         int timeout;
896
897         timeout = 1000;
898         while (!cpu_dead[cpu]) {
899                 if (--timeout == 0) {
900                         printk("CPU %u refused to die!\n", cpu);
901                         break;
902                 }
903                 msleep(1);
904         }
905         cpu_dead[cpu] = 0;
906 }
907
908 #endif /* CONFIG_HOTPLUG_CPU && CONFIG_PP32 */
909
910 /* Core99 Macs (dual G4s and G5s) */
911 struct smp_ops_t core99_smp_ops = {
912         .message_pass   = smp_mpic_message_pass,
913         .probe          = smp_core99_probe,
914         .kick_cpu       = smp_core99_kick_cpu,
915         .setup_cpu      = smp_core99_setup_cpu,
916         .give_timebase  = smp_core99_give_timebase,
917         .take_timebase  = smp_core99_take_timebase,
918 #if defined(CONFIG_HOTPLUG_CPU)
919 # if defined(CONFIG_PPC32)
920         .cpu_disable    = smp_core99_cpu_disable,
921         .cpu_die        = smp_core99_cpu_die,
922 # endif
923 # if defined(CONFIG_PPC64)
924         .cpu_disable    = generic_cpu_disable,
925         .cpu_die        = generic_cpu_die,
926         /* intentionally do *NOT* assign cpu_enable,
927          * the generic code will use kick_cpu then! */
928 # endif
929 #endif
930 };
931
932 void __init pmac_setup_smp(void)
933 {
934         struct device_node *np;
935
936         /* Check for Core99 */
937         np = of_find_node_by_name(NULL, "uni-n");
938         if (!np)
939                 np = of_find_node_by_name(NULL, "u3");
940         if (!np)
941                 np = of_find_node_by_name(NULL, "u4");
942         if (np) {
943                 of_node_put(np);
944                 smp_ops = &core99_smp_ops;
945         }
946 #ifdef CONFIG_PPC32
947         else {
948                 /* We have to set bits in cpu_possible_mask here since the
949                  * secondary CPU(s) aren't in the device tree. Various
950                  * things won't be initialized for CPUs not in the possible
951                  * map, so we really need to fix it up here.
952                  */
953                 int cpu;
954
955                 for (cpu = 1; cpu < 4 && cpu < NR_CPUS; ++cpu)
956                         set_cpu_possible(cpu, true);
957                 smp_ops = &psurge_smp_ops;
958         }
959 #endif /* CONFIG_PPC32 */
960 }
961