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