aboutsummaryrefslogtreecommitdiffstats
path: root/arch/powerpc/platforms/powermac/smp.c
diff options
context:
space:
mode:
Diffstat (limited to 'arch/powerpc/platforms/powermac/smp.c')
-rw-r--r--arch/powerpc/platforms/powermac/smp.c166
1 files changed, 102 insertions, 64 deletions
diff --git a/arch/powerpc/platforms/powermac/smp.c b/arch/powerpc/platforms/powermac/smp.c
index cf1dbe758890..6d4da7b46b41 100644
--- a/arch/powerpc/platforms/powermac/smp.c
+++ b/arch/powerpc/platforms/powermac/smp.c
@@ -64,10 +64,11 @@
64extern void __secondary_start_pmac_0(void); 64extern void __secondary_start_pmac_0(void);
65extern int pmac_pfunc_base_install(void); 65extern int pmac_pfunc_base_install(void);
66 66
67#ifdef CONFIG_PPC32 67static void (*pmac_tb_freeze)(int freeze);
68static u64 timebase;
69static int tb_req;
68 70
69/* Sync flag for HW tb sync */ 71#ifdef CONFIG_PPC32
70static volatile int sec_tb_reset = 0;
71 72
72/* 73/*
73 * Powersurge (old powermac SMP) support. 74 * Powersurge (old powermac SMP) support.
@@ -294,6 +295,9 @@ static int __init smp_psurge_probe(void)
294 psurge_quad_init(); 295 psurge_quad_init();
295 /* All released cards using this HW design have 4 CPUs */ 296 /* All released cards using this HW design have 4 CPUs */
296 ncpus = 4; 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;
297 } else { 301 } else {
298 iounmap(quad_base); 302 iounmap(quad_base);
299 if ((in_8(hhead_base + HHEAD_CONFIG) & 0x02) == 0) { 303 if ((in_8(hhead_base + HHEAD_CONFIG) & 0x02) == 0) {
@@ -308,18 +312,15 @@ static int __init smp_psurge_probe(void)
308 psurge_start = ioremap(PSURGE_START, 4); 312 psurge_start = ioremap(PSURGE_START, 4);
309 psurge_pri_intr = ioremap(PSURGE_PRI_INTR, 4); 313 psurge_pri_intr = ioremap(PSURGE_PRI_INTR, 4);
310 314
311 /* 315 /* This is necessary because OF doesn't know about the
312 * This is necessary because OF doesn't know about the
313 * secondary cpu(s), and thus there aren't nodes in the 316 * secondary cpu(s), and thus there aren't nodes in the
314 * device tree for them, and smp_setup_cpu_maps hasn't 317 * device tree for them, and smp_setup_cpu_maps hasn't
315 * set their bits in cpu_possible_map and cpu_present_map. 318 * set their bits in cpu_present_map.
316 */ 319 */
317 if (ncpus > NR_CPUS) 320 if (ncpus > NR_CPUS)
318 ncpus = NR_CPUS; 321 ncpus = NR_CPUS;
319 for (i = 1; i < ncpus ; ++i) { 322 for (i = 1; i < ncpus ; ++i)
320 cpu_set(i, cpu_present_map); 323 cpu_set(i, cpu_present_map);
321 set_hard_smp_processor_id(i, i);
322 }
323 324
324 if (ppc_md.progress) ppc_md.progress("smp_psurge_probe - done", 0x352); 325 if (ppc_md.progress) ppc_md.progress("smp_psurge_probe - done", 0x352);
325 326
@@ -329,8 +330,14 @@ static int __init smp_psurge_probe(void)
329static void __init smp_psurge_kick_cpu(int nr) 330static void __init smp_psurge_kick_cpu(int nr)
330{ 331{
331 unsigned long start = __pa(__secondary_start_pmac_0) + nr * 8; 332 unsigned long start = __pa(__secondary_start_pmac_0) + nr * 8;
332 unsigned long a; 333 unsigned long a, flags;
333 int i; 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];
334 341
335 /* may need to flush here if secondary bats aren't setup */ 342 /* may need to flush here if secondary bats aren't setup */
336 for (a = KERNELBASE; a < KERNELBASE + 0x800000; a += 32) 343 for (a = KERNELBASE; a < KERNELBASE + 0x800000; a += 32)
@@ -339,47 +346,52 @@ static void __init smp_psurge_kick_cpu(int nr)
339 346
340 if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu", 0x353); 347 if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu", 0x353);
341 348
349 /* This is going to freeze the timeebase, we disable interrupts */
350 local_irq_save(flags);
351
342 out_be32(psurge_start, start); 352 out_be32(psurge_start, start);
343 mb(); 353 mb();
344 354
345 psurge_set_ipi(nr); 355 psurge_set_ipi(nr);
356
346 /* 357 /*
347 * We can't use udelay here because the timebase is now frozen. 358 * We can't use udelay here because the timebase is now frozen.
348 */ 359 */
349 for (i = 0; i < 2000; ++i) 360 for (i = 0; i < 2000; ++i)
350 barrier(); 361 asm volatile("nop" : : : "memory");
351 psurge_clr_ipi(nr); 362 psurge_clr_ipi(nr);
352 363
353 if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu - done", 0x354); 364 /*
354} 365 * Also, because the timebase is frozen, we must not return to the
355 366 * caller which will try to do udelay's etc... Instead, we wait -here-
356/* 367 * for the CPU to callin.
357 * With the dual-cpu powersurge board, the decrementers and timebases 368 */
358 * of both cpus are frozen after the secondary cpu is started up, 369 for (i = 0; i < 100000 && !cpu_callin_map[nr]; ++i) {
359 * until we give the secondary cpu another interrupt. This routine 370 for (j = 1; j < 10000; j++)
360 * uses this to get the timebases synchronized. 371 asm volatile("nop" : : : "memory");
361 * -- paulus. 372 asm volatile("sync" : : : "memory");
362 */ 373 }
363static void __init psurge_dual_sync_tb(int cpu_nr) 374 if (!cpu_callin_map[nr])
364{ 375 goto stuck;
365 int t; 376
366 377 /* And we do the TB sync here too for standard dual CPU cards */
367 set_dec(tb_ticks_per_jiffy); 378 if (psurge_type == PSURGE_DUAL) {
368 /* XXX fixme */ 379 while(!tb_req)
369 set_tb(0, 0); 380 barrier();
370 381 tb_req = 0;
371 if (cpu_nr > 0) { 382 mb();
383 timebase = get_tb();
384 mb();
385 while (timebase)
386 barrier();
372 mb(); 387 mb();
373 sec_tb_reset = 1;
374 return;
375 } 388 }
389 stuck:
390 /* now interrupt the secondary, restarting both TBs */
391 if (psurge_type == PSURGE_DUAL)
392 psurge_set_ipi(1);
376 393
377 /* wait for the secondary to have reset its TB before proceeding */ 394 if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu - done", 0x354);
378 for (t = 10000000; t > 0 && !sec_tb_reset; --t)
379 ;
380
381 /* now interrupt the secondary, starting both TBs */
382 psurge_set_ipi(1);
383} 395}
384 396
385static struct irqaction psurge_irqaction = { 397static struct irqaction psurge_irqaction = {
@@ -390,36 +402,35 @@ static struct irqaction psurge_irqaction = {
390 402
391static void __init smp_psurge_setup_cpu(int cpu_nr) 403static void __init smp_psurge_setup_cpu(int cpu_nr)
392{ 404{
405 if (cpu_nr != 0)
406 return;
393 407
394 if (cpu_nr == 0) { 408 /* reset the entry point so if we get another intr we won't
395 /* If we failed to start the second CPU, we should still 409 * try to startup again */
396 * send it an IPI to start the timebase & DEC or we might 410 out_be32(psurge_start, 0x100);
397 * have them stuck. 411 if (setup_irq(30, &psurge_irqaction))
398 */ 412 printk(KERN_ERR "Couldn't get primary IPI interrupt");
399 if (num_online_cpus() < 2) {
400 if (psurge_type == PSURGE_DUAL)
401 psurge_set_ipi(1);
402 return;
403 }
404 /* reset the entry point so if we get another intr we won't
405 * try to startup again */
406 out_be32(psurge_start, 0x100);
407 if (setup_irq(30, &psurge_irqaction))
408 printk(KERN_ERR "Couldn't get primary IPI interrupt");
409 }
410
411 if (psurge_type == PSURGE_DUAL)
412 psurge_dual_sync_tb(cpu_nr);
413} 413}
414 414
415void __init smp_psurge_take_timebase(void) 415void __init smp_psurge_take_timebase(void)
416{ 416{
417 /* Dummy implementation */ 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);
418} 429}
419 430
420void __init smp_psurge_give_timebase(void) 431void __init smp_psurge_give_timebase(void)
421{ 432{
422 /* Dummy implementation */ 433 /* Nothing to do here */
423} 434}
424 435
425/* PowerSurge-style Macs */ 436/* PowerSurge-style Macs */
@@ -437,9 +448,6 @@ struct smp_ops_t psurge_smp_ops = {
437 * Core 99 and later support 448 * Core 99 and later support
438 */ 449 */
439 450
440static void (*pmac_tb_freeze)(int freeze);
441static u64 timebase;
442static int tb_req;
443 451
444static void smp_core99_give_timebase(void) 452static void smp_core99_give_timebase(void)
445{ 453{
@@ -478,7 +486,6 @@ static void __devinit smp_core99_take_timebase(void)
478 set_tb(timebase >> 32, timebase & 0xffffffff); 486 set_tb(timebase >> 32, timebase & 0xffffffff);
479 timebase = 0; 487 timebase = 0;
480 mb(); 488 mb();
481 set_dec(tb_ticks_per_jiffy/2);
482 489
483 local_irq_restore(flags); 490 local_irq_restore(flags);
484} 491}
@@ -920,3 +927,34 @@ struct smp_ops_t core99_smp_ops = {
920# endif 927# endif
921#endif 928#endif
922}; 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 cpu_set(cpu, cpu_possible_map);
956 smp_ops = &psurge_smp_ops;
957 }
958#endif /* CONFIG_PPC32 */
959}
960