Linux-2.6.12-rc2
[cascardo/linux.git] / arch / ppc / platforms / pmac_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/config.h>
25 #include <linux/kernel.h>
26 #include <linux/sched.h>
27 #include <linux/smp.h>
28 #include <linux/smp_lock.h>
29 #include <linux/interrupt.h>
30 #include <linux/kernel_stat.h>
31 #include <linux/delay.h>
32 #include <linux/init.h>
33 #include <linux/spinlock.h>
34 #include <linux/errno.h>
35 #include <linux/hardirq.h>
36
37 #include <asm/ptrace.h>
38 #include <asm/atomic.h>
39 #include <asm/irq.h>
40 #include <asm/page.h>
41 #include <asm/pgtable.h>
42 #include <asm/sections.h>
43 #include <asm/io.h>
44 #include <asm/prom.h>
45 #include <asm/smp.h>
46 #include <asm/residual.h>
47 #include <asm/machdep.h>
48 #include <asm/pmac_feature.h>
49 #include <asm/time.h>
50 #include <asm/open_pic.h>
51 #include <asm/cacheflush.h>
52 #include <asm/keylargo.h>
53
54 /*
55  * Powersurge (old powermac SMP) support.
56  */
57
58 extern void __secondary_start_psurge(void);
59 extern void __secondary_start_psurge2(void);    /* Temporary horrible hack */
60 extern void __secondary_start_psurge3(void);    /* Temporary horrible hack */
61
62 /* Addresses for powersurge registers */
63 #define HAMMERHEAD_BASE         0xf8000000
64 #define HHEAD_CONFIG            0x90
65 #define HHEAD_SEC_INTR          0xc0
66
67 /* register for interrupting the primary processor on the powersurge */
68 /* N.B. this is actually the ethernet ROM! */
69 #define PSURGE_PRI_INTR         0xf3019000
70
71 /* register for storing the start address for the secondary processor */
72 /* N.B. this is the PCI config space address register for the 1st bridge */
73 #define PSURGE_START            0xf2800000
74
75 /* Daystar/XLR8 4-CPU card */
76 #define PSURGE_QUAD_REG_ADDR    0xf8800000
77
78 #define PSURGE_QUAD_IRQ_SET     0
79 #define PSURGE_QUAD_IRQ_CLR     1
80 #define PSURGE_QUAD_IRQ_PRIMARY 2
81 #define PSURGE_QUAD_CKSTOP_CTL  3
82 #define PSURGE_QUAD_PRIMARY_ARB 4
83 #define PSURGE_QUAD_BOARD_ID    6
84 #define PSURGE_QUAD_WHICH_CPU   7
85 #define PSURGE_QUAD_CKSTOP_RDBK 8
86 #define PSURGE_QUAD_RESET_CTL   11
87
88 #define PSURGE_QUAD_OUT(r, v)   (out_8(quad_base + ((r) << 4) + 4, (v)))
89 #define PSURGE_QUAD_IN(r)       (in_8(quad_base + ((r) << 4) + 4) & 0x0f)
90 #define PSURGE_QUAD_BIS(r, v)   (PSURGE_QUAD_OUT((r), PSURGE_QUAD_IN(r) | (v)))
91 #define PSURGE_QUAD_BIC(r, v)   (PSURGE_QUAD_OUT((r), PSURGE_QUAD_IN(r) & ~(v)))
92
93 /* virtual addresses for the above */
94 static volatile u8 *hhead_base;
95 static volatile u8 *quad_base;
96 static volatile u32 *psurge_pri_intr;
97 static volatile u8 *psurge_sec_intr;
98 static volatile u32 *psurge_start;
99
100 /* values for psurge_type */
101 #define PSURGE_NONE             -1
102 #define PSURGE_DUAL             0
103 #define PSURGE_QUAD_OKEE        1
104 #define PSURGE_QUAD_COTTON      2
105 #define PSURGE_QUAD_ICEGRASS    3
106
107 /* what sort of powersurge board we have */
108 static int psurge_type = PSURGE_NONE;
109
110 /* L2 and L3 cache settings to pass from CPU0 to CPU1 */
111 volatile static long int core99_l2_cache;
112 volatile static long int core99_l3_cache;
113
114 /* Timebase freeze GPIO */
115 static unsigned int core99_tb_gpio;
116
117 /* Sync flag for HW tb sync */
118 static volatile int sec_tb_reset = 0;
119
120 static void __init core99_init_caches(int cpu)
121 {
122         if (!cpu_has_feature(CPU_FTR_L2CR))
123                 return;
124
125         if (cpu == 0) {
126                 core99_l2_cache = _get_L2CR();
127                 printk("CPU0: L2CR is %lx\n", core99_l2_cache);
128         } else {
129                 printk("CPU%d: L2CR was %lx\n", cpu, _get_L2CR());
130                 _set_L2CR(0);
131                 _set_L2CR(core99_l2_cache);
132                 printk("CPU%d: L2CR set to %lx\n", cpu, core99_l2_cache);
133         }
134
135         if (!cpu_has_feature(CPU_FTR_L3CR))
136                 return;
137
138         if (cpu == 0){
139                 core99_l3_cache = _get_L3CR();
140                 printk("CPU0: L3CR is %lx\n", core99_l3_cache);
141         } else {
142                 printk("CPU%d: L3CR was %lx\n", cpu, _get_L3CR());
143                 _set_L3CR(0);
144                 _set_L3CR(core99_l3_cache);
145                 printk("CPU%d: L3CR set to %lx\n", cpu, core99_l3_cache);
146         }
147 }
148
149 /*
150  * Set and clear IPIs for powersurge.
151  */
152 static inline void psurge_set_ipi(int cpu)
153 {
154         if (psurge_type == PSURGE_NONE)
155                 return;
156         if (cpu == 0)
157                 in_be32(psurge_pri_intr);
158         else if (psurge_type == PSURGE_DUAL)
159                 out_8(psurge_sec_intr, 0);
160         else
161                 PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_SET, 1 << cpu);
162 }
163
164 static inline void psurge_clr_ipi(int cpu)
165 {
166         if (cpu > 0) {
167                 switch(psurge_type) {
168                 case PSURGE_DUAL:
169                         out_8(psurge_sec_intr, ~0);
170                 case PSURGE_NONE:
171                         break;
172                 default:
173                         PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_CLR, 1 << cpu);
174                 }
175         }
176 }
177
178 /*
179  * On powersurge (old SMP powermac architecture) we don't have
180  * separate IPIs for separate messages like openpic does.  Instead
181  * we have a bitmap for each processor, where a 1 bit means that
182  * the corresponding message is pending for that processor.
183  * Ideally each cpu's entry would be in a different cache line.
184  *  -- paulus.
185  */
186 static unsigned long psurge_smp_message[NR_CPUS];
187
188 void __pmac psurge_smp_message_recv(struct pt_regs *regs)
189 {
190         int cpu = smp_processor_id();
191         int msg;
192
193         /* clear interrupt */
194         psurge_clr_ipi(cpu);
195
196         if (num_online_cpus() < 2)
197                 return;
198
199         /* make sure there is a message there */
200         for (msg = 0; msg < 4; msg++)
201                 if (test_and_clear_bit(msg, &psurge_smp_message[cpu]))
202                         smp_message_recv(msg, regs);
203 }
204
205 irqreturn_t __pmac psurge_primary_intr(int irq, void *d, struct pt_regs *regs)
206 {
207         psurge_smp_message_recv(regs);
208         return IRQ_HANDLED;
209 }
210
211 static void __pmac smp_psurge_message_pass(int target, int msg, unsigned long data,
212                                            int wait)
213 {
214         int i;
215
216         if (num_online_cpus() < 2)
217                 return;
218
219         for (i = 0; i < NR_CPUS; i++) {
220                 if (!cpu_online(i))
221                         continue;
222                 if (target == MSG_ALL
223                     || (target == MSG_ALL_BUT_SELF && i != smp_processor_id())
224                     || target == i) {
225                         set_bit(msg, &psurge_smp_message[i]);
226                         psurge_set_ipi(i);
227                 }
228         }
229 }
230
231 /*
232  * Determine a quad card presence. We read the board ID register, we
233  * force the data bus to change to something else, and we read it again.
234  * It it's stable, then the register probably exist (ugh !)
235  */
236 static int __init psurge_quad_probe(void)
237 {
238         int type;
239         unsigned int i;
240
241         type = PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID);
242         if (type < PSURGE_QUAD_OKEE || type > PSURGE_QUAD_ICEGRASS
243             || type != PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID))
244                 return PSURGE_DUAL;
245
246         /* looks OK, try a slightly more rigorous test */
247         /* bogus is not necessarily cacheline-aligned,
248            though I don't suppose that really matters.  -- paulus */
249         for (i = 0; i < 100; i++) {
250                 volatile u32 bogus[8];
251                 bogus[(0+i)%8] = 0x00000000;
252                 bogus[(1+i)%8] = 0x55555555;
253                 bogus[(2+i)%8] = 0xFFFFFFFF;
254                 bogus[(3+i)%8] = 0xAAAAAAAA;
255                 bogus[(4+i)%8] = 0x33333333;
256                 bogus[(5+i)%8] = 0xCCCCCCCC;
257                 bogus[(6+i)%8] = 0xCCCCCCCC;
258                 bogus[(7+i)%8] = 0x33333333;
259                 wmb();
260                 asm volatile("dcbf 0,%0" : : "r" (bogus) : "memory");
261                 mb();
262                 if (type != PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID))
263                         return PSURGE_DUAL;
264         }
265         return type;
266 }
267
268 static void __init psurge_quad_init(void)
269 {
270         int procbits;
271
272         if (ppc_md.progress) ppc_md.progress("psurge_quad_init", 0x351);
273         procbits = ~PSURGE_QUAD_IN(PSURGE_QUAD_WHICH_CPU);
274         if (psurge_type == PSURGE_QUAD_ICEGRASS)
275                 PSURGE_QUAD_BIS(PSURGE_QUAD_RESET_CTL, procbits);
276         else
277                 PSURGE_QUAD_BIC(PSURGE_QUAD_CKSTOP_CTL, procbits);
278         mdelay(33);
279         out_8(psurge_sec_intr, ~0);
280         PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_CLR, procbits);
281         PSURGE_QUAD_BIS(PSURGE_QUAD_RESET_CTL, procbits);
282         if (psurge_type != PSURGE_QUAD_ICEGRASS)
283                 PSURGE_QUAD_BIS(PSURGE_QUAD_CKSTOP_CTL, procbits);
284         PSURGE_QUAD_BIC(PSURGE_QUAD_PRIMARY_ARB, procbits);
285         mdelay(33);
286         PSURGE_QUAD_BIC(PSURGE_QUAD_RESET_CTL, procbits);
287         mdelay(33);
288         PSURGE_QUAD_BIS(PSURGE_QUAD_PRIMARY_ARB, procbits);
289         mdelay(33);
290 }
291
292 static int __init smp_psurge_probe(void)
293 {
294         int i, ncpus;
295
296         /* We don't do SMP on the PPC601 -- paulus */
297         if (PVR_VER(mfspr(SPRN_PVR)) == 1)
298                 return 1;
299
300         /*
301          * The powersurge cpu board can be used in the generation
302          * of powermacs that have a socket for an upgradeable cpu card,
303          * including the 7500, 8500, 9500, 9600.
304          * The device tree doesn't tell you if you have 2 cpus because
305          * OF doesn't know anything about the 2nd processor.
306          * Instead we look for magic bits in magic registers,
307          * in the hammerhead memory controller in the case of the
308          * dual-cpu powersurge board.  -- paulus.
309          */
310         if (find_devices("hammerhead") == NULL)
311                 return 1;
312
313         hhead_base = ioremap(HAMMERHEAD_BASE, 0x800);
314         quad_base = ioremap(PSURGE_QUAD_REG_ADDR, 1024);
315         psurge_sec_intr = hhead_base + HHEAD_SEC_INTR;
316
317         psurge_type = psurge_quad_probe();
318         if (psurge_type != PSURGE_DUAL) {
319                 psurge_quad_init();
320                 /* All released cards using this HW design have 4 CPUs */
321                 ncpus = 4;
322         } else {
323                 iounmap((void *) quad_base);
324                 if ((in_8(hhead_base + HHEAD_CONFIG) & 0x02) == 0) {
325                         /* not a dual-cpu card */
326                         iounmap((void *) hhead_base);
327                         psurge_type = PSURGE_NONE;
328                         return 1;
329                 }
330                 ncpus = 2;
331         }
332
333         psurge_start = ioremap(PSURGE_START, 4);
334         psurge_pri_intr = ioremap(PSURGE_PRI_INTR, 4);
335
336         /* this is not actually strictly necessary -- paulus. */
337         for (i = 1; i < ncpus; ++i)
338                 smp_hw_index[i] = i;
339
340         if (ppc_md.progress) ppc_md.progress("smp_psurge_probe - done", 0x352);
341
342         return ncpus;
343 }
344
345 static void __init smp_psurge_kick_cpu(int nr)
346 {
347         void (*start)(void) = __secondary_start_psurge;
348         unsigned long a;
349
350         /* may need to flush here if secondary bats aren't setup */
351         for (a = KERNELBASE; a < KERNELBASE + 0x800000; a += 32)
352                 asm volatile("dcbf 0,%0" : : "r" (a) : "memory");
353         asm volatile("sync");
354
355         if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu", 0x353);
356
357         /* setup entry point of secondary processor */
358         switch (nr) {
359         case 2:
360                 start = __secondary_start_psurge2;
361                 break;
362         case 3:
363                 start = __secondary_start_psurge3;
364                 break;
365         }
366
367         out_be32(psurge_start, __pa(start));
368         mb();
369
370         psurge_set_ipi(nr);
371         udelay(10);
372         psurge_clr_ipi(nr);
373
374         if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu - done", 0x354);
375 }
376
377 /*
378  * With the dual-cpu powersurge board, the decrementers and timebases
379  * of both cpus are frozen after the secondary cpu is started up,
380  * until we give the secondary cpu another interrupt.  This routine
381  * uses this to get the timebases synchronized.
382  *  -- paulus.
383  */
384 static void __init psurge_dual_sync_tb(int cpu_nr)
385 {
386         int t;
387
388         set_dec(tb_ticks_per_jiffy);
389         set_tb(0, 0);
390         last_jiffy_stamp(cpu_nr) = 0;
391
392         if (cpu_nr > 0) {
393                 mb();
394                 sec_tb_reset = 1;
395                 return;
396         }
397
398         /* wait for the secondary to have reset its TB before proceeding */
399         for (t = 10000000; t > 0 && !sec_tb_reset; --t)
400                 ;
401
402         /* now interrupt the secondary, starting both TBs */
403         psurge_set_ipi(1);
404
405         smp_tb_synchronized = 1;
406 }
407
408 static struct irqaction psurge_irqaction = {
409         .handler = psurge_primary_intr,
410         .flags = SA_INTERRUPT,
411         .mask = CPU_MASK_NONE,
412         .name = "primary IPI",
413 };
414
415 static void __init smp_psurge_setup_cpu(int cpu_nr)
416 {
417
418         if (cpu_nr == 0) {
419                 /* If we failed to start the second CPU, we should still
420                  * send it an IPI to start the timebase & DEC or we might
421                  * have them stuck.
422                  */
423                 if (num_online_cpus() < 2) {
424                         if (psurge_type == PSURGE_DUAL)
425                                 psurge_set_ipi(1);
426                         return;
427                 }
428                 /* reset the entry point so if we get another intr we won't
429                  * try to startup again */
430                 out_be32(psurge_start, 0x100);
431                 if (setup_irq(30, &psurge_irqaction))
432                         printk(KERN_ERR "Couldn't get primary IPI interrupt");
433         }
434
435         if (psurge_type == PSURGE_DUAL)
436                 psurge_dual_sync_tb(cpu_nr);
437 }
438
439 void __init smp_psurge_take_timebase(void)
440 {
441         /* Dummy implementation */
442 }
443
444 void __init smp_psurge_give_timebase(void)
445 {
446         /* Dummy implementation */
447 }
448
449 static int __init smp_core99_probe(void)
450 {
451 #ifdef CONFIG_6xx
452         extern int powersave_nap;
453 #endif
454         struct device_node *cpus, *firstcpu;
455         int i, ncpus = 0, boot_cpu = -1;
456         u32 *tbprop;
457
458         if (ppc_md.progress) ppc_md.progress("smp_core99_probe", 0x345);
459         cpus = firstcpu = find_type_devices("cpu");
460         while(cpus != NULL) {
461                 u32 *regprop = (u32 *)get_property(cpus, "reg", NULL);
462                 char *stateprop = (char *)get_property(cpus, "state", NULL);
463                 if (regprop != NULL && stateprop != NULL &&
464                     !strncmp(stateprop, "running", 7))
465                         boot_cpu = *regprop;
466                 ++ncpus;
467                 cpus = cpus->next;
468         }
469         if (boot_cpu == -1)
470                 printk(KERN_WARNING "Couldn't detect boot CPU !\n");
471         if (boot_cpu != 0)
472                 printk(KERN_WARNING "Boot CPU is %d, unsupported setup !\n", boot_cpu);
473
474         if (machine_is_compatible("MacRISC4")) {
475                 extern struct smp_ops_t core99_smp_ops;
476
477                 core99_smp_ops.take_timebase = smp_generic_take_timebase;
478                 core99_smp_ops.give_timebase = smp_generic_give_timebase;
479         } else {
480                 if (firstcpu != NULL)
481                         tbprop = (u32 *)get_property(firstcpu, "timebase-enable", NULL);
482                 if (tbprop)
483                         core99_tb_gpio = *tbprop;
484                 else
485                         core99_tb_gpio = KL_GPIO_TB_ENABLE;
486         }
487
488         if (ncpus > 1) {
489                 openpic_request_IPIs();
490                 for (i = 1; i < ncpus; ++i)
491                         smp_hw_index[i] = i;
492 #ifdef CONFIG_6xx
493                 powersave_nap = 0;
494 #endif
495                 core99_init_caches(0);
496         }
497
498         return ncpus;
499 }
500
501 static void __init smp_core99_kick_cpu(int nr)
502 {
503         unsigned long save_vector, new_vector;
504         unsigned long flags;
505
506         volatile unsigned long *vector
507                  = ((volatile unsigned long *)(KERNELBASE+0x100));
508         if (nr < 1 || nr > 3)
509                 return;
510         if (ppc_md.progress) ppc_md.progress("smp_core99_kick_cpu", 0x346);
511
512         local_irq_save(flags);
513         local_irq_disable();
514
515         /* Save reset vector */
516         save_vector = *vector;
517
518         /* Setup fake reset vector that does    
519          *   b __secondary_start_psurge - KERNELBASE
520          */
521         switch(nr) {
522                 case 1:
523                         new_vector = (unsigned long)__secondary_start_psurge;
524                         break;
525                 case 2:
526                         new_vector = (unsigned long)__secondary_start_psurge2;
527                         break;
528                 case 3:
529                         new_vector = (unsigned long)__secondary_start_psurge3;
530                         break;
531         }
532         *vector = 0x48000002 + new_vector - KERNELBASE;
533
534         /* flush data cache and inval instruction cache */
535         flush_icache_range((unsigned long) vector, (unsigned long) vector + 4);
536
537         /* Put some life in our friend */
538         pmac_call_feature(PMAC_FTR_RESET_CPU, NULL, nr, 0);
539
540         /* FIXME: We wait a bit for the CPU to take the exception, I should
541          * instead wait for the entry code to set something for me. Well,
542          * ideally, all that crap will be done in prom.c and the CPU left
543          * in a RAM-based wait loop like CHRP.
544          */
545         mdelay(1);
546
547         /* Restore our exception vector */
548         *vector = save_vector;
549         flush_icache_range((unsigned long) vector, (unsigned long) vector + 4);
550
551         local_irq_restore(flags);
552         if (ppc_md.progress) ppc_md.progress("smp_core99_kick_cpu done", 0x347);
553 }
554
555 static void __init smp_core99_setup_cpu(int cpu_nr)
556 {
557         /* Setup L2/L3 */
558         if (cpu_nr != 0)
559                 core99_init_caches(cpu_nr);
560
561         /* Setup openpic */
562         do_openpic_setup_cpu();
563
564         if (cpu_nr == 0) {
565 #ifdef CONFIG_POWER4
566                 extern void g5_phy_disable_cpu1(void);
567
568                 /* If we didn't start the second CPU, we must take
569                  * it off the bus
570                  */
571                 if (machine_is_compatible("MacRISC4") &&
572                     num_online_cpus() < 2)              
573                         g5_phy_disable_cpu1();
574 #endif /* CONFIG_POWER4 */
575                 if (ppc_md.progress) ppc_md.progress("core99_setup_cpu 0 done", 0x349);
576         }
577 }
578
579 void __init smp_core99_take_timebase(void)
580 {
581         /* Secondary processor "takes" the timebase by freezing
582          * it, resetting its local TB and telling CPU 0 to go on
583          */
584         pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 4);
585         pmac_call_feature(PMAC_FTR_READ_GPIO, NULL, core99_tb_gpio, 0);
586         mb();
587
588         set_dec(tb_ticks_per_jiffy);
589         set_tb(0, 0);
590         last_jiffy_stamp(smp_processor_id()) = 0;
591
592         mb();
593         sec_tb_reset = 1;
594 }
595
596 void __init smp_core99_give_timebase(void)
597 {
598         unsigned int t;
599
600         /* Primary processor waits for secondary to have frozen
601          * the timebase, resets local TB, and kick timebase again
602          */
603         /* wait for the secondary to have reset its TB before proceeding */
604         for (t = 1000; t > 0 && !sec_tb_reset; --t)
605                 udelay(1000);
606         if (t == 0)
607                 printk(KERN_WARNING "Timeout waiting sync on second CPU\n");
608
609         set_dec(tb_ticks_per_jiffy);
610         set_tb(0, 0);
611         last_jiffy_stamp(smp_processor_id()) = 0;
612         mb();
613
614         /* Now, restart the timebase by leaving the GPIO to an open collector */
615         pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 0);
616         pmac_call_feature(PMAC_FTR_READ_GPIO, NULL, core99_tb_gpio, 0);
617
618         smp_tb_synchronized = 1;
619 }
620
621
622 /* PowerSurge-style Macs */
623 struct smp_ops_t psurge_smp_ops __pmacdata = {
624         .message_pass   = smp_psurge_message_pass,
625         .probe          = smp_psurge_probe,
626         .kick_cpu       = smp_psurge_kick_cpu,
627         .setup_cpu      = smp_psurge_setup_cpu,
628         .give_timebase  = smp_psurge_give_timebase,
629         .take_timebase  = smp_psurge_take_timebase,
630 };
631
632 /* Core99 Macs (dual G4s) */
633 struct smp_ops_t core99_smp_ops __pmacdata = {
634         .message_pass   = smp_openpic_message_pass,
635         .probe          = smp_core99_probe,
636         .kick_cpu       = smp_core99_kick_cpu,
637         .setup_cpu      = smp_core99_setup_cpu,
638         .give_timebase  = smp_core99_give_timebase,
639         .take_timebase  = smp_core99_take_timebase,
640 };