Remove obsolete #include <linux/config.h>
[linux-2.6.git] / arch / parisc / kernel / smp.c
1 /*
2 ** SMP Support
3 **
4 ** Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
5 ** Copyright (C) 1999 David Mosberger-Tang <davidm@hpl.hp.com>
6 ** Copyright (C) 2001,2004 Grant Grundler <grundler@parisc-linux.org>
7 ** 
8 ** Lots of stuff stolen from arch/alpha/kernel/smp.c
9 ** ...and then parisc stole from arch/ia64/kernel/smp.c. Thanks David! :^)
10 **
11 ** Thanks to John Curry and Ullas Ponnadi. I learned alot from their work.
12 ** -grant (1/12/2001)
13 **
14 **      This program is free software; you can redistribute it and/or modify
15 **      it under the terms of the GNU General Public License as published by
16 **      the Free Software Foundation; either version 2 of the License, or
17 **      (at your option) any later version.
18 */
19 #undef ENTRY_SYS_CPUS   /* syscall support for iCOD-like functionality */
20
21
22 #include <linux/types.h>
23 #include <linux/spinlock.h>
24 #include <linux/slab.h>
25
26 #include <linux/kernel.h>
27 #include <linux/module.h>
28 #include <linux/sched.h>
29 #include <linux/init.h>
30 #include <linux/interrupt.h>
31 #include <linux/smp.h>
32 #include <linux/kernel_stat.h>
33 #include <linux/mm.h>
34 #include <linux/delay.h>
35 #include <linux/bitops.h>
36
37 #include <asm/system.h>
38 #include <asm/atomic.h>
39 #include <asm/current.h>
40 #include <asm/delay.h>
41 #include <asm/tlbflush.h>
42
43 #include <asm/io.h>
44 #include <asm/irq.h>            /* for CPU_IRQ_REGION and friends */
45 #include <asm/mmu_context.h>
46 #include <asm/page.h>
47 #include <asm/pgtable.h>
48 #include <asm/pgalloc.h>
49 #include <asm/processor.h>
50 #include <asm/ptrace.h>
51 #include <asm/unistd.h>
52 #include <asm/cacheflush.h>
53
54 #define kDEBUG 0
55
56 DEFINE_SPINLOCK(smp_lock);
57
58 volatile struct task_struct *smp_init_current_idle_task;
59
60 static volatile int cpu_now_booting __read_mostly = 0;  /* track which CPU is booting */
61
62 static int parisc_max_cpus __read_mostly = 1;
63
64 /* online cpus are ones that we've managed to bring up completely
65  * possible cpus are all valid cpu 
66  * present cpus are all detected cpu
67  *
68  * On startup we bring up the "possible" cpus. Since we discover
69  * CPUs later, we add them as hotplug, so the possible cpu mask is
70  * empty in the beginning.
71  */
72
73 cpumask_t cpu_online_map   __read_mostly = CPU_MASK_NONE;       /* Bitmap of online CPUs */
74 cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL;        /* Bitmap of Present CPUs */
75
76 EXPORT_SYMBOL(cpu_online_map);
77 EXPORT_SYMBOL(cpu_possible_map);
78
79
80 struct smp_call_struct {
81         void (*func) (void *info);
82         void *info;
83         long wait;
84         atomic_t unstarted_count;
85         atomic_t unfinished_count;
86 };
87 static volatile struct smp_call_struct *smp_call_function_data;
88
89 enum ipi_message_type {
90         IPI_NOP=0,
91         IPI_RESCHEDULE=1,
92         IPI_CALL_FUNC,
93         IPI_CPU_START,
94         IPI_CPU_STOP,
95         IPI_CPU_TEST
96 };
97
98
99 /********** SMP inter processor interrupt and communication routines */
100
101 #undef PER_CPU_IRQ_REGION
102 #ifdef PER_CPU_IRQ_REGION
103 /* XXX REVISIT Ignore for now.
104 **    *May* need this "hook" to register IPI handler
105 **    once we have perCPU ExtIntr switch tables.
106 */
107 static void
108 ipi_init(int cpuid)
109 {
110
111         /* If CPU is present ... */
112 #ifdef ENTRY_SYS_CPUS
113         /* *and* running (not stopped) ... */
114 #error iCOD support wants state checked here.
115 #endif
116
117 #error verify IRQ_OFFSET(IPI_IRQ) is ipi_interrupt() in new IRQ region
118
119         if(cpu_online(cpuid) )
120         {
121                 switch_to_idle_task(current);
122         }
123
124         return;
125 }
126 #endif
127
128
129 /*
130 ** Yoink this CPU from the runnable list... 
131 **
132 */
133 static void
134 halt_processor(void) 
135 {
136 #ifdef ENTRY_SYS_CPUS
137 #error halt_processor() needs rework
138 /*
139 ** o migrate I/O interrupts off this CPU.
140 ** o leave IPI enabled - __cli() will disable IPI.
141 ** o leave CPU in online map - just change the state
142 */
143         cpu_data[this_cpu].state = STATE_STOPPED;
144         mark_bh(IPI_BH);
145 #else
146         /* REVISIT : redirect I/O Interrupts to another CPU? */
147         /* REVISIT : does PM *know* this CPU isn't available? */
148         cpu_clear(smp_processor_id(), cpu_online_map);
149         local_irq_disable();
150         for (;;)
151                 ;
152 #endif
153 }
154
155
156 irqreturn_t
157 ipi_interrupt(int irq, void *dev_id, struct pt_regs *regs) 
158 {
159         int this_cpu = smp_processor_id();
160         struct cpuinfo_parisc *p = &cpu_data[this_cpu];
161         unsigned long ops;
162         unsigned long flags;
163
164         /* Count this now; we may make a call that never returns. */
165         p->ipi_count++;
166
167         mb();   /* Order interrupt and bit testing. */
168
169         for (;;) {
170                 spin_lock_irqsave(&(p->lock),flags);
171                 ops = p->pending_ipi;
172                 p->pending_ipi = 0;
173                 spin_unlock_irqrestore(&(p->lock),flags);
174
175                 mb(); /* Order bit clearing and data access. */
176
177                 if (!ops)
178                     break;
179
180                 while (ops) {
181                         unsigned long which = ffz(~ops);
182
183                         ops &= ~(1 << which);
184
185                         switch (which) {
186                         case IPI_NOP:
187 #if (kDEBUG>=100)
188                                 printk(KERN_DEBUG "CPU%d IPI_NOP\n",this_cpu);
189 #endif /* kDEBUG */
190                                 break;
191                                 
192                         case IPI_RESCHEDULE:
193 #if (kDEBUG>=100)
194                                 printk(KERN_DEBUG "CPU%d IPI_RESCHEDULE\n",this_cpu);
195 #endif /* kDEBUG */
196                                 /*
197                                  * Reschedule callback.  Everything to be
198                                  * done is done by the interrupt return path.
199                                  */
200                                 break;
201
202                         case IPI_CALL_FUNC:
203 #if (kDEBUG>=100)
204                                 printk(KERN_DEBUG "CPU%d IPI_CALL_FUNC\n",this_cpu);
205 #endif /* kDEBUG */
206                                 {
207                                         volatile struct smp_call_struct *data;
208                                         void (*func)(void *info);
209                                         void *info;
210                                         int wait;
211
212                                         data = smp_call_function_data;
213                                         func = data->func;
214                                         info = data->info;
215                                         wait = data->wait;
216
217                                         mb();
218                                         atomic_dec ((atomic_t *)&data->unstarted_count);
219
220                                         /* At this point, *data can't
221                                          * be relied upon.
222                                          */
223
224                                         (*func)(info);
225
226                                         /* Notify the sending CPU that the
227                                          * task is done.
228                                          */
229                                         mb();
230                                         if (wait)
231                                                 atomic_dec ((atomic_t *)&data->unfinished_count);
232                                 }
233                                 break;
234
235                         case IPI_CPU_START:
236 #if (kDEBUG>=100)
237                                 printk(KERN_DEBUG "CPU%d IPI_CPU_START\n",this_cpu);
238 #endif /* kDEBUG */
239 #ifdef ENTRY_SYS_CPUS
240                                 p->state = STATE_RUNNING;
241 #endif
242                                 break;
243
244                         case IPI_CPU_STOP:
245 #if (kDEBUG>=100)
246                                 printk(KERN_DEBUG "CPU%d IPI_CPU_STOP\n",this_cpu);
247 #endif /* kDEBUG */
248 #ifdef ENTRY_SYS_CPUS
249 #else
250                                 halt_processor();
251 #endif
252                                 break;
253
254                         case IPI_CPU_TEST:
255 #if (kDEBUG>=100)
256                                 printk(KERN_DEBUG "CPU%d is alive!\n",this_cpu);
257 #endif /* kDEBUG */
258                                 break;
259
260                         default:
261                                 printk(KERN_CRIT "Unknown IPI num on CPU%d: %lu\n",
262                                         this_cpu, which);
263                                 return IRQ_NONE;
264                         } /* Switch */
265                 } /* while (ops) */
266         }
267         return IRQ_HANDLED;
268 }
269
270
271 static inline void
272 ipi_send(int cpu, enum ipi_message_type op)
273 {
274         struct cpuinfo_parisc *p = &cpu_data[cpu];
275         unsigned long flags;
276
277         spin_lock_irqsave(&(p->lock),flags);
278         p->pending_ipi |= 1 << op;
279         gsc_writel(IPI_IRQ - CPU_IRQ_BASE, cpu_data[cpu].hpa);
280         spin_unlock_irqrestore(&(p->lock),flags);
281 }
282
283
284 static inline void
285 send_IPI_single(int dest_cpu, enum ipi_message_type op)
286 {
287         if (dest_cpu == NO_PROC_ID) {
288                 BUG();
289                 return;
290         }
291
292         ipi_send(dest_cpu, op);
293 }
294
295 static inline void
296 send_IPI_allbutself(enum ipi_message_type op)
297 {
298         int i;
299         
300         for_each_online_cpu(i) {
301                 if (i != smp_processor_id())
302                         send_IPI_single(i, op);
303         }
304 }
305
306
307 inline void 
308 smp_send_stop(void)     { send_IPI_allbutself(IPI_CPU_STOP); }
309
310 static inline void
311 smp_send_start(void)    { send_IPI_allbutself(IPI_CPU_START); }
312
313 void 
314 smp_send_reschedule(int cpu) { send_IPI_single(cpu, IPI_RESCHEDULE); }
315
316 void
317 smp_send_all_nop(void)
318 {
319         send_IPI_allbutself(IPI_NOP);
320 }
321
322
323 /**
324  * Run a function on all other CPUs.
325  *  <func>      The function to run. This must be fast and non-blocking.
326  *  <info>      An arbitrary pointer to pass to the function.
327  *  <retry>     If true, keep retrying until ready.
328  *  <wait>      If true, wait until function has completed on other CPUs.
329  *  [RETURNS]   0 on success, else a negative status code.
330  *
331  * Does not return until remote CPUs are nearly ready to execute <func>
332  * or have executed.
333  */
334
335 int
336 smp_call_function (void (*func) (void *info), void *info, int retry, int wait)
337 {
338         struct smp_call_struct data;
339         unsigned long timeout;
340         static DEFINE_SPINLOCK(lock);
341         int retries = 0;
342
343         if (num_online_cpus() < 2)
344                 return 0;
345
346         /* Can deadlock when called with interrupts disabled */
347         WARN_ON(irqs_disabled());
348
349         /* can also deadlock if IPIs are disabled */
350         WARN_ON((get_eiem() & (1UL<<(CPU_IRQ_MAX - IPI_IRQ))) == 0);
351
352         
353         data.func = func;
354         data.info = info;
355         data.wait = wait;
356         atomic_set(&data.unstarted_count, num_online_cpus() - 1);
357         atomic_set(&data.unfinished_count, num_online_cpus() - 1);
358
359         if (retry) {
360                 spin_lock (&lock);
361                 while (smp_call_function_data != 0)
362                         barrier();
363         }
364         else {
365                 spin_lock (&lock);
366                 if (smp_call_function_data) {
367                         spin_unlock (&lock);
368                         return -EBUSY;
369                 }
370         }
371
372         smp_call_function_data = &data;
373         spin_unlock (&lock);
374         
375         /*  Send a message to all other CPUs and wait for them to respond  */
376         send_IPI_allbutself(IPI_CALL_FUNC);
377
378  retry:
379         /*  Wait for response  */
380         timeout = jiffies + HZ;
381         while ( (atomic_read (&data.unstarted_count) > 0) &&
382                 time_before (jiffies, timeout) )
383                 barrier ();
384
385         if (atomic_read (&data.unstarted_count) > 0) {
386                 printk(KERN_CRIT "SMP CALL FUNCTION TIMED OUT! (cpu=%d), try %d\n",
387                       smp_processor_id(), ++retries);
388                 goto retry;
389         }
390         /* We either got one or timed out. Release the lock */
391
392         mb();
393         smp_call_function_data = NULL;
394
395         while (wait && atomic_read (&data.unfinished_count) > 0)
396                         barrier ();
397
398         return 0;
399 }
400
401 EXPORT_SYMBOL(smp_call_function);
402
403 /*
404  * Flush all other CPU's tlb and then mine.  Do this with on_each_cpu()
405  * as we want to ensure all TLB's flushed before proceeding.
406  */
407
408 void
409 smp_flush_tlb_all(void)
410 {
411         on_each_cpu(flush_tlb_all_local, NULL, 1, 1);
412 }
413
414
415 void 
416 smp_do_timer(struct pt_regs *regs)
417 {
418         int cpu = smp_processor_id();
419         struct cpuinfo_parisc *data = &cpu_data[cpu];
420
421         if (!--data->prof_counter) {
422                 data->prof_counter = data->prof_multiplier;
423                 update_process_times(user_mode(regs));
424         }
425 }
426
427 /*
428  * Called by secondaries to update state and initialize CPU registers.
429  */
430 static void __init
431 smp_cpu_init(int cpunum)
432 {
433         extern int init_per_cpu(int);  /* arch/parisc/kernel/setup.c */
434         extern void init_IRQ(void);    /* arch/parisc/kernel/irq.c */
435
436         /* Set modes and Enable floating point coprocessor */
437         (void) init_per_cpu(cpunum);
438
439         disable_sr_hashing();
440
441         mb();
442
443         /* Well, support 2.4 linux scheme as well. */
444         if (cpu_test_and_set(cpunum, cpu_online_map))
445         {
446                 extern void machine_halt(void); /* arch/parisc.../process.c */
447
448                 printk(KERN_CRIT "CPU#%d already initialized!\n", cpunum);
449                 machine_halt();
450         }  
451
452         /* Initialise the idle task for this CPU */
453         atomic_inc(&init_mm.mm_count);
454         current->active_mm = &init_mm;
455         if(current->mm)
456                 BUG();
457         enter_lazy_tlb(&init_mm, current);
458
459         init_IRQ();   /* make sure no IRQ's are enabled or pending */
460 }
461
462
463 /*
464  * Slaves start using C here. Indirectly called from smp_slave_stext.
465  * Do what start_kernel() and main() do for boot strap processor (aka monarch)
466  */
467 void __init smp_callin(void)
468 {
469         int slave_id = cpu_now_booting;
470 #if 0
471         void *istack;
472 #endif
473
474         smp_cpu_init(slave_id);
475         preempt_disable();
476
477 #if 0   /* NOT WORKING YET - see entry.S */
478         istack = (void *)__get_free_pages(GFP_KERNEL,ISTACK_ORDER);
479         if (istack == NULL) {
480             printk(KERN_CRIT "Failed to allocate interrupt stack for cpu %d\n",slave_id);
481             BUG();
482         }
483         mtctl(istack,31);
484 #endif
485
486         flush_cache_all_local(); /* start with known state */
487         flush_tlb_all_local(NULL);
488
489         local_irq_enable();  /* Interrupts have been off until now */
490
491         cpu_idle();      /* Wait for timer to schedule some work */
492
493         /* NOTREACHED */
494         panic("smp_callin() AAAAaaaaahhhh....\n");
495 }
496
497 /*
498  * Bring one cpu online.
499  */
500 int __init smp_boot_one_cpu(int cpuid)
501 {
502         struct task_struct *idle;
503         long timeout;
504
505         /* 
506          * Create an idle task for this CPU.  Note the address wed* give 
507          * to kernel_thread is irrelevant -- it's going to start
508          * where OS_BOOT_RENDEVZ vector in SAL says to start.  But
509          * this gets all the other task-y sort of data structures set
510          * up like we wish.   We need to pull the just created idle task 
511          * off the run queue and stuff it into the init_tasks[] array.  
512          * Sheesh . . .
513          */
514
515         idle = fork_idle(cpuid);
516         if (IS_ERR(idle))
517                 panic("SMP: fork failed for CPU:%d", cpuid);
518
519         task_thread_info(idle)->cpu = cpuid;
520
521         /* Let _start know what logical CPU we're booting
522         ** (offset into init_tasks[],cpu_data[])
523         */
524         cpu_now_booting = cpuid;
525
526         /* 
527         ** boot strap code needs to know the task address since
528         ** it also contains the process stack.
529         */
530         smp_init_current_idle_task = idle ;
531         mb();
532
533         printk("Releasing cpu %d now, hpa=%lx\n", cpuid, cpu_data[cpuid].hpa);
534
535         /*
536         ** This gets PDC to release the CPU from a very tight loop.
537         **
538         ** From the PA-RISC 2.0 Firmware Architecture Reference Specification:
539         ** "The MEM_RENDEZ vector specifies the location of OS_RENDEZ which 
540         ** is executed after receiving the rendezvous signal (an interrupt to 
541         ** EIR{0}). MEM_RENDEZ is valid only when it is nonzero and the 
542         ** contents of memory are valid."
543         */
544         gsc_writel(TIMER_IRQ - CPU_IRQ_BASE, cpu_data[cpuid].hpa);
545         mb();
546
547         /* 
548          * OK, wait a bit for that CPU to finish staggering about. 
549          * Slave will set a bit when it reaches smp_cpu_init().
550          * Once the "monarch CPU" sees the bit change, it can move on.
551          */
552         for (timeout = 0; timeout < 10000; timeout++) {
553                 if(cpu_online(cpuid)) {
554                         /* Which implies Slave has started up */
555                         cpu_now_booting = 0;
556                         smp_init_current_idle_task = NULL;
557                         goto alive ;
558                 }
559                 udelay(100);
560                 barrier();
561         }
562
563         put_task_struct(idle);
564         idle = NULL;
565
566         printk(KERN_CRIT "SMP: CPU:%d is stuck.\n", cpuid);
567         return -1;
568
569 alive:
570         /* Remember the Slave data */
571 #if (kDEBUG>=100)
572         printk(KERN_DEBUG "SMP: CPU:%d came alive after %ld _us\n",
573                 cpuid, timeout * 100);
574 #endif /* kDEBUG */
575 #ifdef ENTRY_SYS_CPUS
576         cpu_data[cpuid].state = STATE_RUNNING;
577 #endif
578         return 0;
579 }
580
581 void __devinit smp_prepare_boot_cpu(void)
582 {
583         int bootstrap_processor=cpu_data[0].cpuid;      /* CPU ID of BSP */
584
585 #ifdef ENTRY_SYS_CPUS
586         cpu_data[0].state = STATE_RUNNING;
587 #endif
588
589         /* Setup BSP mappings */
590         printk("SMP: bootstrap CPU ID is %d\n",bootstrap_processor);
591
592         cpu_set(bootstrap_processor, cpu_online_map);
593         cpu_set(bootstrap_processor, cpu_present_map);
594 }
595
596
597
598 /*
599 ** inventory.c:do_inventory() hasn't yet been run and thus we
600 ** don't 'discover' the additional CPU's until later.
601 */
602 void __init smp_prepare_cpus(unsigned int max_cpus)
603 {
604         cpus_clear(cpu_present_map);
605         cpu_set(0, cpu_present_map);
606
607         parisc_max_cpus = max_cpus;
608         if (!max_cpus)
609                 printk(KERN_INFO "SMP mode deactivated.\n");
610 }
611
612
613 void smp_cpus_done(unsigned int cpu_max)
614 {
615         return;
616 }
617
618
619 int __devinit __cpu_up(unsigned int cpu)
620 {
621         if (cpu != 0 && cpu < parisc_max_cpus)
622                 smp_boot_one_cpu(cpu);
623
624         return cpu_online(cpu) ? 0 : -ENOSYS;
625 }
626
627
628
629 #ifdef ENTRY_SYS_CPUS
630 /* Code goes along with:
631 **    entry.s:        ENTRY_NAME(sys_cpus)   / * 215, for cpu stat * /
632 */
633 int sys_cpus(int argc, char **argv)
634 {
635         int i,j=0;
636         extern int current_pid(int cpu);
637
638         if( argc > 2 ) {
639                 printk("sys_cpus:Only one argument supported\n");
640                 return (-1);
641         }
642         if ( argc == 1 ){
643         
644 #ifdef DUMP_MORE_STATE
645                 for_each_online_cpu(i) {
646                         int cpus_per_line = 4;
647
648                         if (j++ % cpus_per_line)
649                                 printk(" %3d",i);
650                         else
651                                 printk("\n %3d",i);
652                 }
653                 printk("\n"); 
654 #else
655                 printk("\n 0\n"); 
656 #endif
657         } else if((argc==2) && !(strcmp(argv[1],"-l"))) {
658                 printk("\nCPUSTATE  TASK CPUNUM CPUID HARDCPU(HPA)\n");
659 #ifdef DUMP_MORE_STATE
660                 for_each_online_cpu(i) {
661                         if (cpu_data[i].cpuid != NO_PROC_ID) {
662                                 switch(cpu_data[i].state) {
663                                         case STATE_RENDEZVOUS:
664                                                 printk("RENDEZVS ");
665                                                 break;
666                                         case STATE_RUNNING:
667                                                 printk((current_pid(i)!=0) ? "RUNNING  " : "IDLING   ");
668                                                 break;
669                                         case STATE_STOPPED:
670                                                 printk("STOPPED  ");
671                                                 break;
672                                         case STATE_HALTED:
673                                                 printk("HALTED   ");
674                                                 break;
675                                         default:
676                                                 printk("%08x?", cpu_data[i].state);
677                                                 break;
678                                 }
679                                 if(cpu_online(i)) {
680                                         printk(" %4d",current_pid(i));
681                                 }       
682                                 printk(" %6d",cpu_number_map(i));
683                                 printk(" %5d",i);
684                                 printk(" 0x%lx\n",cpu_data[i].hpa);
685                         }       
686                 }
687 #else
688                 printk("\n%s  %4d      0     0 --------",
689                         (current->pid)?"RUNNING ": "IDLING  ",current->pid); 
690 #endif
691         } else if ((argc==2) && !(strcmp(argv[1],"-s"))) { 
692 #ifdef DUMP_MORE_STATE
693                 printk("\nCPUSTATE   CPUID\n");
694                 for_each_online_cpu(i) {
695                         if (cpu_data[i].cpuid != NO_PROC_ID) {
696                                 switch(cpu_data[i].state) {
697                                         case STATE_RENDEZVOUS:
698                                                 printk("RENDEZVS");break;
699                                         case STATE_RUNNING:
700                                                 printk((current_pid(i)!=0) ? "RUNNING " : "IDLING");
701                                                 break;
702                                         case STATE_STOPPED:
703                                                 printk("STOPPED ");break;
704                                         case STATE_HALTED:
705                                                 printk("HALTED  ");break;
706                                         default:
707                                 }
708                                 printk("  %5d\n",i);
709                         }       
710                 }
711 #else
712                 printk("\n%s    CPU0",(current->pid==0)?"RUNNING ":"IDLING  "); 
713 #endif
714         } else {
715                 printk("sys_cpus:Unknown request\n");
716                 return (-1);
717         }
718         return 0;
719 }
720 #endif /* ENTRY_SYS_CPUS */
721
722 #ifdef CONFIG_PROC_FS
723 int __init
724 setup_profiling_timer(unsigned int multiplier)
725 {
726         return -EINVAL;
727 }
728 #endif