Pull sgi-drivers-makefile-cleanup into release branch
[linux-2.6] / arch / sparc / kernel / sun4d_smp.c
1 /* sun4d_smp.c: Sparc SS1000/SC2000 SMP support.
2  *
3  * Copyright (C) 1998 Jakub Jelinek (jj@sunsite.mff.cuni.cz)
4  *
5  * Based on sun4m's smp.c, which is:
6  * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu)
7  */
8
9 #include <asm/head.h>
10
11 #include <linux/kernel.h>
12 #include <linux/sched.h>
13 #include <linux/threads.h>
14 #include <linux/smp.h>
15 #include <linux/smp_lock.h>
16 #include <linux/interrupt.h>
17 #include <linux/kernel_stat.h>
18 #include <linux/init.h>
19 #include <linux/spinlock.h>
20 #include <linux/mm.h>
21 #include <linux/swap.h>
22 #include <linux/profile.h>
23
24 #include <asm/ptrace.h>
25 #include <asm/atomic.h>
26
27 #include <asm/delay.h>
28 #include <asm/irq.h>
29 #include <asm/page.h>
30 #include <asm/pgalloc.h>
31 #include <asm/pgtable.h>
32 #include <asm/oplib.h>
33 #include <asm/sbus.h>
34 #include <asm/sbi.h>
35 #include <asm/tlbflush.h>
36 #include <asm/cacheflush.h>
37 #include <asm/cpudata.h>
38
39 #define IRQ_CROSS_CALL          15
40
41 extern ctxd_t *srmmu_ctx_table_phys;
42
43 extern void calibrate_delay(void);
44
45 extern volatile int smp_processors_ready;
46 extern int smp_num_cpus;
47 static int smp_highest_cpu;
48 extern volatile unsigned long cpu_callin_map[NR_CPUS];
49 extern struct cpuinfo_sparc cpu_data[NR_CPUS];
50 extern unsigned char boot_cpu_id;
51 extern int smp_activated;
52 extern volatile int __cpu_number_map[NR_CPUS];
53 extern volatile int __cpu_logical_map[NR_CPUS];
54 extern volatile unsigned long ipi_count;
55 extern volatile int smp_process_available;
56 extern volatile int smp_commenced;
57 extern int __smp4d_processor_id(void);
58
59 /* #define SMP_DEBUG */
60
61 #ifdef SMP_DEBUG
62 #define SMP_PRINTK(x)   printk x
63 #else
64 #define SMP_PRINTK(x)
65 #endif
66
67 static inline unsigned long swap(volatile unsigned long *ptr, unsigned long val)
68 {
69         __asm__ __volatile__("swap [%1], %0\n\t" :
70                              "=&r" (val), "=&r" (ptr) :
71                              "0" (val), "1" (ptr));
72         return val;
73 }
74
75 static void smp_setup_percpu_timer(void);
76 extern void cpu_probe(void);
77 extern void sun4d_distribute_irqs(void);
78
79 void __init smp4d_callin(void)
80 {
81         int cpuid = hard_smp4d_processor_id();
82         extern spinlock_t sun4d_imsk_lock;
83         unsigned long flags;
84         
85         /* Show we are alive */
86         cpu_leds[cpuid] = 0x6;
87         show_leds(cpuid);
88
89         /* Enable level15 interrupt, disable level14 interrupt for now */
90         cc_set_imsk((cc_get_imsk() & ~0x8000) | 0x4000);
91
92         local_flush_cache_all();
93         local_flush_tlb_all();
94
95         /*
96          * Unblock the master CPU _only_ when the scheduler state
97          * of all secondary CPUs will be up-to-date, so after
98          * the SMP initialization the master will be just allowed
99          * to call the scheduler code.
100          */
101         /* Get our local ticker going. */
102         smp_setup_percpu_timer();
103
104         calibrate_delay();
105         smp_store_cpu_info(cpuid);
106         local_flush_cache_all();
107         local_flush_tlb_all();
108
109         /* Allow master to continue. */
110         swap((unsigned long *)&cpu_callin_map[cpuid], 1);
111         local_flush_cache_all();
112         local_flush_tlb_all();
113         
114         cpu_probe();
115
116         while((unsigned long)current_set[cpuid] < PAGE_OFFSET)
117                 barrier();
118                 
119         while(current_set[cpuid]->cpu != cpuid)
120                 barrier();
121                 
122         /* Fix idle thread fields. */
123         __asm__ __volatile__("ld [%0], %%g6\n\t"
124                              : : "r" (&current_set[cpuid])
125                              : "memory" /* paranoid */);
126
127         cpu_leds[cpuid] = 0x9;
128         show_leds(cpuid);
129         
130         /* Attach to the address space of init_task. */
131         atomic_inc(&init_mm.mm_count);
132         current->active_mm = &init_mm;
133
134         local_flush_cache_all();
135         local_flush_tlb_all();
136         
137         local_irq_enable();     /* We don't allow PIL 14 yet */
138         
139         while(!smp_commenced)
140                 barrier();
141
142         spin_lock_irqsave(&sun4d_imsk_lock, flags);
143         cc_set_imsk(cc_get_imsk() & ~0x4000); /* Allow PIL 14 as well */
144         spin_unlock_irqrestore(&sun4d_imsk_lock, flags);
145 }
146
147 extern void init_IRQ(void);
148 extern void cpu_panic(void);
149
150 /*
151  *      Cycle through the processors asking the PROM to start each one.
152  */
153  
154 extern struct linux_prom_registers smp_penguin_ctable;
155 extern unsigned long trapbase_cpu1[];
156 extern unsigned long trapbase_cpu2[];
157 extern unsigned long trapbase_cpu3[];
158
159 void __init smp4d_boot_cpus(void)
160 {
161         int cpucount = 0;
162         int i, mid;
163
164         printk("Entering SMP Mode...\n");
165         
166         if (boot_cpu_id)
167                 current_set[0] = NULL;
168
169         local_irq_enable();
170         cpus_clear(cpu_present_map);
171
172         /* XXX This whole thing has to go.  See sparc64. */
173         for (i = 0; !cpu_find_by_instance(i, NULL, &mid); i++)
174                 cpu_set(mid, cpu_present_map);
175         SMP_PRINTK(("cpu_present_map %08lx\n", cpus_addr(cpu_present_map)[0]));
176         for(i=0; i < NR_CPUS; i++)
177                 __cpu_number_map[i] = -1;
178         for(i=0; i < NR_CPUS; i++)
179                 __cpu_logical_map[i] = -1;
180         __cpu_number_map[boot_cpu_id] = 0;
181         __cpu_logical_map[0] = boot_cpu_id;
182         current_thread_info()->cpu = boot_cpu_id;
183         smp_store_cpu_info(boot_cpu_id);
184         smp_setup_percpu_timer();
185         local_flush_cache_all();
186         if (cpu_find_by_instance(1, NULL, NULL))
187                 return;  /* Not an MP box. */
188         SMP_PRINTK(("Iterating over CPUs\n"));
189         for(i = 0; i < NR_CPUS; i++) {
190                 if(i == boot_cpu_id)
191                         continue;
192
193                 if (cpu_isset(i, cpu_present_map)) {
194                         extern unsigned long sun4d_cpu_startup;
195                         unsigned long *entry = &sun4d_cpu_startup;
196                         struct task_struct *p;
197                         int timeout;
198                         int no;
199
200                         /* Cook up an idler for this guy. */
201                         p = fork_idle(i);
202                         cpucount++;
203                         current_set[i] = task_thread_info(p);
204                         for (no = 0; !cpu_find_by_instance(no, NULL, &mid)
205                                      && mid != i; no++) ;
206
207                         /*
208                          * Initialize the contexts table
209                          * Since the call to prom_startcpu() trashes the structure,
210                          * we need to re-initialize it for each cpu
211                          */
212                         smp_penguin_ctable.which_io = 0;
213                         smp_penguin_ctable.phys_addr = (unsigned int) srmmu_ctx_table_phys;
214                         smp_penguin_ctable.reg_size = 0;
215
216                         /* whirrr, whirrr, whirrrrrrrrr... */
217                         SMP_PRINTK(("Starting CPU %d at %p task %d node %08x\n", i, entry, cpucount, cpu_data(no).prom_node));
218                         local_flush_cache_all();
219                         prom_startcpu(cpu_data(no).prom_node,
220                                       &smp_penguin_ctable, 0, (char *)entry);
221                                       
222                         SMP_PRINTK(("prom_startcpu returned :)\n"));
223
224                         /* wheee... it's going... */
225                         for(timeout = 0; timeout < 10000; timeout++) {
226                                 if(cpu_callin_map[i])
227                                         break;
228                                 udelay(200);
229                         }
230                         
231                         if(cpu_callin_map[i]) {
232                                 /* Another "Red Snapper". */
233                                 __cpu_number_map[i] = cpucount;
234                                 __cpu_logical_map[cpucount] = i;
235                         } else {
236                                 cpucount--;
237                                 printk("Processor %d is stuck.\n", i);
238                         }
239                 }
240                 if(!(cpu_callin_map[i])) {
241                         cpu_clear(i, cpu_present_map);
242                         __cpu_number_map[i] = -1;
243                 }
244         }
245         local_flush_cache_all();
246         if(cpucount == 0) {
247                 printk("Error: only one Processor found.\n");
248                 cpu_present_map = cpumask_of_cpu(hard_smp4d_processor_id());
249         } else {
250                 unsigned long bogosum = 0;
251                 
252                 for(i = 0; i < NR_CPUS; i++) {
253                         if (cpu_isset(i, cpu_present_map)) {
254                                 bogosum += cpu_data(i).udelay_val;
255                                 smp_highest_cpu = i;
256                         }
257                 }
258                 SMP_PRINTK(("Total of %d Processors activated (%lu.%02lu BogoMIPS).\n", cpucount + 1, bogosum/(500000/HZ), (bogosum/(5000/HZ))%100));
259                 printk("Total of %d Processors activated (%lu.%02lu BogoMIPS).\n",
260                        cpucount + 1,
261                        bogosum/(500000/HZ),
262                        (bogosum/(5000/HZ))%100);
263                 smp_activated = 1;
264                 smp_num_cpus = cpucount + 1;
265         }
266
267         /* Free unneeded trap tables */
268         ClearPageReserved(virt_to_page(trapbase_cpu1));
269         set_page_count(virt_to_page(trapbase_cpu1), 1);
270         free_page((unsigned long)trapbase_cpu1);
271         totalram_pages++;
272         num_physpages++;
273
274         ClearPageReserved(virt_to_page(trapbase_cpu2));
275         set_page_count(virt_to_page(trapbase_cpu2), 1);
276         free_page((unsigned long)trapbase_cpu2);
277         totalram_pages++;
278         num_physpages++;
279
280         ClearPageReserved(virt_to_page(trapbase_cpu3));
281         set_page_count(virt_to_page(trapbase_cpu3), 1);
282         free_page((unsigned long)trapbase_cpu3);
283         totalram_pages++;
284         num_physpages++;
285
286         /* Ok, they are spinning and ready to go. */
287         smp_processors_ready = 1;
288         sun4d_distribute_irqs();
289 }
290
291 static struct smp_funcall {
292         smpfunc_t func;
293         unsigned long arg1;
294         unsigned long arg2;
295         unsigned long arg3;
296         unsigned long arg4;
297         unsigned long arg5;
298         unsigned char processors_in[NR_CPUS];  /* Set when ipi entered. */
299         unsigned char processors_out[NR_CPUS]; /* Set when ipi exited. */
300 } ccall_info __attribute__((aligned(8)));
301
302 static DEFINE_SPINLOCK(cross_call_lock);
303
304 /* Cross calls must be serialized, at least currently. */
305 void smp4d_cross_call(smpfunc_t func, unsigned long arg1, unsigned long arg2,
306                     unsigned long arg3, unsigned long arg4, unsigned long arg5)
307 {
308         if(smp_processors_ready) {
309                 register int high = smp_highest_cpu;
310                 unsigned long flags;
311
312                 spin_lock_irqsave(&cross_call_lock, flags);
313
314                 {
315                         /* If you make changes here, make sure gcc generates proper code... */
316                         register smpfunc_t f asm("i0") = func;
317                         register unsigned long a1 asm("i1") = arg1;
318                         register unsigned long a2 asm("i2") = arg2;
319                         register unsigned long a3 asm("i3") = arg3;
320                         register unsigned long a4 asm("i4") = arg4;
321                         register unsigned long a5 asm("i5") = arg5;
322
323                         __asm__ __volatile__(
324                                 "std %0, [%6]\n\t"
325                                 "std %2, [%6 + 8]\n\t"
326                                 "std %4, [%6 + 16]\n\t" : :
327                                 "r"(f), "r"(a1), "r"(a2), "r"(a3), "r"(a4), "r"(a5),
328                                 "r" (&ccall_info.func));
329                 }
330
331                 /* Init receive/complete mapping, plus fire the IPI's off. */
332                 {
333                         cpumask_t mask;
334                         register int i;
335
336                         mask = cpumask_of_cpu(hard_smp4d_processor_id());
337                         cpus_andnot(mask, cpu_present_map, mask);
338                         for(i = 0; i <= high; i++) {
339                                 if (cpu_isset(i, mask)) {
340                                         ccall_info.processors_in[i] = 0;
341                                         ccall_info.processors_out[i] = 0;
342                                         sun4d_send_ipi(i, IRQ_CROSS_CALL);
343                                 }
344                         }
345                 }
346
347                 {
348                         register int i;
349
350                         i = 0;
351                         do {
352                                 while(!ccall_info.processors_in[i])
353                                         barrier();
354                         } while(++i <= high);
355
356                         i = 0;
357                         do {
358                                 while(!ccall_info.processors_out[i])
359                                         barrier();
360                         } while(++i <= high);
361                 }
362
363                 spin_unlock_irqrestore(&cross_call_lock, flags);
364         }
365 }
366
367 /* Running cross calls. */
368 void smp4d_cross_call_irq(void)
369 {
370         int i = hard_smp4d_processor_id();
371
372         ccall_info.processors_in[i] = 1;
373         ccall_info.func(ccall_info.arg1, ccall_info.arg2, ccall_info.arg3,
374                         ccall_info.arg4, ccall_info.arg5);
375         ccall_info.processors_out[i] = 1;
376 }
377
378 static int smp4d_stop_cpu_sender;
379
380 static void smp4d_stop_cpu(void)
381 {
382         int me = hard_smp4d_processor_id();
383         
384         if (me != smp4d_stop_cpu_sender)
385                 while(1) barrier();
386 }
387
388 /* Cross calls, in order to work efficiently and atomically do all
389  * the message passing work themselves, only stopcpu and reschedule
390  * messages come through here.
391  */
392 void smp4d_message_pass(int target, int msg, unsigned long data, int wait)
393 {
394         int me = hard_smp4d_processor_id();
395
396         SMP_PRINTK(("smp4d_message_pass %d %d %08lx %d\n", target, msg, data, wait));
397         if (msg == MSG_STOP_CPU && target == MSG_ALL_BUT_SELF) {
398                 unsigned long flags;
399                 static DEFINE_SPINLOCK(stop_cpu_lock);
400                 spin_lock_irqsave(&stop_cpu_lock, flags);
401                 smp4d_stop_cpu_sender = me;
402                 smp4d_cross_call((smpfunc_t)smp4d_stop_cpu, 0, 0, 0, 0, 0);
403                 spin_unlock_irqrestore(&stop_cpu_lock, flags);
404         }
405         printk("Yeeee, trying to send SMP msg(%d) to %d on cpu %d\n", msg, target, me);
406         panic("Bogon SMP message pass.");
407 }
408
409 void smp4d_percpu_timer_interrupt(struct pt_regs *regs)
410 {
411         int cpu = hard_smp4d_processor_id();
412         static int cpu_tick[NR_CPUS];
413         static char led_mask[] = { 0xe, 0xd, 0xb, 0x7, 0xb, 0xd };
414
415         bw_get_prof_limit(cpu); 
416         bw_clear_intr_mask(0, 1);       /* INTR_TABLE[0] & 1 is Profile IRQ */
417
418         cpu_tick[cpu]++;
419         if (!(cpu_tick[cpu] & 15)) {
420                 if (cpu_tick[cpu] == 0x60)
421                         cpu_tick[cpu] = 0;
422                 cpu_leds[cpu] = led_mask[cpu_tick[cpu] >> 4];
423                 show_leds(cpu);
424         }
425
426         profile_tick(CPU_PROFILING, regs);
427
428         if(!--prof_counter(cpu)) {
429                 int user = user_mode(regs);
430
431                 irq_enter();
432                 update_process_times(user);
433                 irq_exit();
434
435                 prof_counter(cpu) = prof_multiplier(cpu);
436         }
437 }
438
439 extern unsigned int lvl14_resolution;
440
441 static void __init smp_setup_percpu_timer(void)
442 {
443         int cpu = hard_smp4d_processor_id();
444
445         prof_counter(cpu) = prof_multiplier(cpu) = 1;
446         load_profile_irq(cpu, lvl14_resolution);
447 }
448
449 void __init smp4d_blackbox_id(unsigned *addr)
450 {
451         int rd = *addr & 0x3e000000;
452         
453         addr[0] = 0xc0800800 | rd;              /* lda [%g0] ASI_M_VIKING_TMP1, reg */
454         addr[1] = 0x01000000;                   /* nop */
455         addr[2] = 0x01000000;                   /* nop */
456 }
457
458 void __init smp4d_blackbox_current(unsigned *addr)
459 {
460         int rd = *addr & 0x3e000000;
461         
462         addr[0] = 0xc0800800 | rd;              /* lda [%g0] ASI_M_VIKING_TMP1, reg */
463         addr[2] = 0x81282002 | rd | (rd >> 11); /* sll reg, 2, reg */
464         addr[4] = 0x01000000;                   /* nop */
465 }
466
467 void __init sun4d_init_smp(void)
468 {
469         int i;
470         extern unsigned int t_nmi[], linux_trap_ipi15_sun4d[], linux_trap_ipi15_sun4m[];
471
472         /* Patch ipi15 trap table */
473         t_nmi[1] = t_nmi[1] + (linux_trap_ipi15_sun4d - linux_trap_ipi15_sun4m);
474         
475         /* And set btfixup... */
476         BTFIXUPSET_BLACKBOX(hard_smp_processor_id, smp4d_blackbox_id);
477         BTFIXUPSET_BLACKBOX(load_current, smp4d_blackbox_current);
478         BTFIXUPSET_CALL(smp_cross_call, smp4d_cross_call, BTFIXUPCALL_NORM);
479         BTFIXUPSET_CALL(smp_message_pass, smp4d_message_pass, BTFIXUPCALL_NORM);
480         BTFIXUPSET_CALL(__hard_smp_processor_id, __smp4d_processor_id, BTFIXUPCALL_NORM);
481         
482         for (i = 0; i < NR_CPUS; i++) {
483                 ccall_info.processors_in[i] = 1;
484                 ccall_info.processors_out[i] = 1;
485         }
486 }