Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6
[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/interrupt.h>
16 #include <linux/kernel_stat.h>
17 #include <linux/init.h>
18 #include <linux/spinlock.h>
19 #include <linux/mm.h>
20 #include <linux/swap.h>
21 #include <linux/profile.h>
22
23 #include <asm/ptrace.h>
24 #include <asm/atomic.h>
25 #include <asm/irq_regs.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 #include "irq.h"
40 #define IRQ_CROSS_CALL          15
41
42 extern ctxd_t *srmmu_ctx_table_phys;
43
44 extern void calibrate_delay(void);
45
46 static volatile int smp_processors_ready = 0;
47 static int smp_highest_cpu;
48 extern volatile unsigned long cpu_callin_map[NR_CPUS];
49 extern cpuinfo_sparc cpu_data[NR_CPUS];
50 extern unsigned char boot_cpu_id;
51 extern volatile int smp_process_available;
52
53 extern cpumask_t smp_commenced_mask;
54
55 extern int __smp4d_processor_id(void);
56
57 /* #define SMP_DEBUG */
58
59 #ifdef SMP_DEBUG
60 #define SMP_PRINTK(x)   printk x
61 #else
62 #define SMP_PRINTK(x)
63 #endif
64
65 static inline unsigned long swap(volatile unsigned long *ptr, unsigned long val)
66 {
67         __asm__ __volatile__("swap [%1], %0\n\t" :
68                              "=&r" (val), "=&r" (ptr) :
69                              "0" (val), "1" (ptr));
70         return val;
71 }
72
73 static void smp_setup_percpu_timer(void);
74 extern void cpu_probe(void);
75 extern void sun4d_distribute_irqs(void);
76
77 void __init smp4d_callin(void)
78 {
79         int cpuid = hard_smp4d_processor_id();
80         extern spinlock_t sun4d_imsk_lock;
81         unsigned long flags;
82         
83         /* Show we are alive */
84         cpu_leds[cpuid] = 0x6;
85         show_leds(cpuid);
86
87         /* Enable level15 interrupt, disable level14 interrupt for now */
88         cc_set_imsk((cc_get_imsk() & ~0x8000) | 0x4000);
89
90         local_flush_cache_all();
91         local_flush_tlb_all();
92
93         /*
94          * Unblock the master CPU _only_ when the scheduler state
95          * of all secondary CPUs will be up-to-date, so after
96          * the SMP initialization the master will be just allowed
97          * to call the scheduler code.
98          */
99         /* Get our local ticker going. */
100         smp_setup_percpu_timer();
101
102         calibrate_delay();
103         smp_store_cpu_info(cpuid);
104         local_flush_cache_all();
105         local_flush_tlb_all();
106
107         /* Allow master to continue. */
108         swap((unsigned long *)&cpu_callin_map[cpuid], 1);
109         local_flush_cache_all();
110         local_flush_tlb_all();
111         
112         cpu_probe();
113
114         while((unsigned long)current_set[cpuid] < PAGE_OFFSET)
115                 barrier();
116                 
117         while(current_set[cpuid]->cpu != cpuid)
118                 barrier();
119                 
120         /* Fix idle thread fields. */
121         __asm__ __volatile__("ld [%0], %%g6\n\t"
122                              : : "r" (&current_set[cpuid])
123                              : "memory" /* paranoid */);
124
125         cpu_leds[cpuid] = 0x9;
126         show_leds(cpuid);
127         
128         /* Attach to the address space of init_task. */
129         atomic_inc(&init_mm.mm_count);
130         current->active_mm = &init_mm;
131
132         local_flush_cache_all();
133         local_flush_tlb_all();
134         
135         local_irq_enable();     /* We don't allow PIL 14 yet */
136         
137         while (!cpu_isset(cpuid, smp_commenced_mask))
138                 barrier();
139
140         spin_lock_irqsave(&sun4d_imsk_lock, flags);
141         cc_set_imsk(cc_get_imsk() & ~0x4000); /* Allow PIL 14 as well */
142         spin_unlock_irqrestore(&sun4d_imsk_lock, flags);
143         cpu_set(cpuid, cpu_online_map);
144
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         if (boot_cpu_id)
162                 current_set[0] = NULL;
163         smp_setup_percpu_timer();
164         local_flush_cache_all();
165 }
166
167 int __cpuinit smp4d_boot_one_cpu(int i)
168 {
169                         extern unsigned long sun4d_cpu_startup;
170                         unsigned long *entry = &sun4d_cpu_startup;
171                         struct task_struct *p;
172                         int timeout;
173                         int cpu_node;
174
175                         cpu_find_by_instance(i, &cpu_node,NULL);
176                         /* Cook up an idler for this guy. */
177                         p = fork_idle(i);
178                         current_set[i] = task_thread_info(p);
179
180                         /*
181                          * Initialize the contexts table
182                          * Since the call to prom_startcpu() trashes the structure,
183                          * we need to re-initialize it for each cpu
184                          */
185                         smp_penguin_ctable.which_io = 0;
186                         smp_penguin_ctable.phys_addr = (unsigned int) srmmu_ctx_table_phys;
187                         smp_penguin_ctable.reg_size = 0;
188
189                         /* whirrr, whirrr, whirrrrrrrrr... */
190                         SMP_PRINTK(("Starting CPU %d at %p \n", i, entry));
191                         local_flush_cache_all();
192                         prom_startcpu(cpu_node,
193                                       &smp_penguin_ctable, 0, (char *)entry);
194                                       
195                         SMP_PRINTK(("prom_startcpu returned :)\n"));
196
197                         /* wheee... it's going... */
198                         for(timeout = 0; timeout < 10000; timeout++) {
199                                 if(cpu_callin_map[i])
200                                         break;
201                                 udelay(200);
202                         }
203                         
204         if (!(cpu_callin_map[i])) {
205                 printk("Processor %d is stuck.\n", i);
206                 return -ENODEV;
207
208         }
209         local_flush_cache_all();
210         return 0;
211 }
212
213 void __init smp4d_smp_done(void)
214 {
215         int i, first;
216         int *prev;
217
218         /* setup cpu list for irq rotation */
219         first = 0;
220         prev = &first;
221         for (i = 0; i < NR_CPUS; i++)
222                 if (cpu_online(i)) {
223                         *prev = i;
224                         prev = &cpu_data(i).next;
225                 }
226         *prev = first;
227         local_flush_cache_all();
228
229         /* Free unneeded trap tables */
230         ClearPageReserved(virt_to_page(trapbase_cpu1));
231         init_page_count(virt_to_page(trapbase_cpu1));
232         free_page((unsigned long)trapbase_cpu1);
233         totalram_pages++;
234         num_physpages++;
235
236         ClearPageReserved(virt_to_page(trapbase_cpu2));
237         init_page_count(virt_to_page(trapbase_cpu2));
238         free_page((unsigned long)trapbase_cpu2);
239         totalram_pages++;
240         num_physpages++;
241
242         ClearPageReserved(virt_to_page(trapbase_cpu3));
243         init_page_count(virt_to_page(trapbase_cpu3));
244         free_page((unsigned long)trapbase_cpu3);
245         totalram_pages++;
246         num_physpages++;
247
248         /* Ok, they are spinning and ready to go. */
249         smp_processors_ready = 1;
250         sun4d_distribute_irqs();
251 }
252
253 static struct smp_funcall {
254         smpfunc_t func;
255         unsigned long arg1;
256         unsigned long arg2;
257         unsigned long arg3;
258         unsigned long arg4;
259         unsigned long arg5;
260         unsigned char processors_in[NR_CPUS];  /* Set when ipi entered. */
261         unsigned char processors_out[NR_CPUS]; /* Set when ipi exited. */
262 } ccall_info __attribute__((aligned(8)));
263
264 static DEFINE_SPINLOCK(cross_call_lock);
265
266 /* Cross calls must be serialized, at least currently. */
267 void smp4d_cross_call(smpfunc_t func, unsigned long arg1, unsigned long arg2,
268                     unsigned long arg3, unsigned long arg4, unsigned long arg5)
269 {
270         if(smp_processors_ready) {
271                 register int high = smp_highest_cpu;
272                 unsigned long flags;
273
274                 spin_lock_irqsave(&cross_call_lock, flags);
275
276                 {
277                         /* If you make changes here, make sure gcc generates proper code... */
278                         register smpfunc_t f asm("i0") = func;
279                         register unsigned long a1 asm("i1") = arg1;
280                         register unsigned long a2 asm("i2") = arg2;
281                         register unsigned long a3 asm("i3") = arg3;
282                         register unsigned long a4 asm("i4") = arg4;
283                         register unsigned long a5 asm("i5") = arg5;
284
285                         __asm__ __volatile__(
286                                 "std %0, [%6]\n\t"
287                                 "std %2, [%6 + 8]\n\t"
288                                 "std %4, [%6 + 16]\n\t" : :
289                                 "r"(f), "r"(a1), "r"(a2), "r"(a3), "r"(a4), "r"(a5),
290                                 "r" (&ccall_info.func));
291                 }
292
293                 /* Init receive/complete mapping, plus fire the IPI's off. */
294                 {
295                         cpumask_t mask;
296                         register int i;
297
298                         mask = cpumask_of_cpu(hard_smp4d_processor_id());
299                         cpus_andnot(mask, cpu_online_map, mask);
300                         for(i = 0; i <= high; i++) {
301                                 if (cpu_isset(i, mask)) {
302                                         ccall_info.processors_in[i] = 0;
303                                         ccall_info.processors_out[i] = 0;
304                                         sun4d_send_ipi(i, IRQ_CROSS_CALL);
305                                 }
306                         }
307                 }
308
309                 {
310                         register int i;
311
312                         i = 0;
313                         do {
314                                 while(!ccall_info.processors_in[i])
315                                         barrier();
316                         } while(++i <= high);
317
318                         i = 0;
319                         do {
320                                 while(!ccall_info.processors_out[i])
321                                         barrier();
322                         } while(++i <= high);
323                 }
324
325                 spin_unlock_irqrestore(&cross_call_lock, flags);
326         }
327 }
328
329 /* Running cross calls. */
330 void smp4d_cross_call_irq(void)
331 {
332         int i = hard_smp4d_processor_id();
333
334         ccall_info.processors_in[i] = 1;
335         ccall_info.func(ccall_info.arg1, ccall_info.arg2, ccall_info.arg3,
336                         ccall_info.arg4, ccall_info.arg5);
337         ccall_info.processors_out[i] = 1;
338 }
339
340 static int smp4d_stop_cpu_sender;
341
342 static void smp4d_stop_cpu(void)
343 {
344         int me = hard_smp4d_processor_id();
345         
346         if (me != smp4d_stop_cpu_sender)
347                 while(1) barrier();
348 }
349
350 /* Cross calls, in order to work efficiently and atomically do all
351  * the message passing work themselves, only stopcpu and reschedule
352  * messages come through here.
353  */
354 void smp4d_message_pass(int target, int msg, unsigned long data, int wait)
355 {
356         int me = hard_smp4d_processor_id();
357
358         SMP_PRINTK(("smp4d_message_pass %d %d %08lx %d\n", target, msg, data, wait));
359         if (msg == MSG_STOP_CPU && target == MSG_ALL_BUT_SELF) {
360                 unsigned long flags;
361                 static DEFINE_SPINLOCK(stop_cpu_lock);
362                 spin_lock_irqsave(&stop_cpu_lock, flags);
363                 smp4d_stop_cpu_sender = me;
364                 smp4d_cross_call((smpfunc_t)smp4d_stop_cpu, 0, 0, 0, 0, 0);
365                 spin_unlock_irqrestore(&stop_cpu_lock, flags);
366         }
367         printk("Yeeee, trying to send SMP msg(%d) to %d on cpu %d\n", msg, target, me);
368         panic("Bogon SMP message pass.");
369 }
370
371 void smp4d_percpu_timer_interrupt(struct pt_regs *regs)
372 {
373         struct pt_regs *old_regs;
374         int cpu = hard_smp4d_processor_id();
375         static int cpu_tick[NR_CPUS];
376         static char led_mask[] = { 0xe, 0xd, 0xb, 0x7, 0xb, 0xd };
377
378         old_regs = set_irq_regs(regs);
379         bw_get_prof_limit(cpu); 
380         bw_clear_intr_mask(0, 1);       /* INTR_TABLE[0] & 1 is Profile IRQ */
381
382         cpu_tick[cpu]++;
383         if (!(cpu_tick[cpu] & 15)) {
384                 if (cpu_tick[cpu] == 0x60)
385                         cpu_tick[cpu] = 0;
386                 cpu_leds[cpu] = led_mask[cpu_tick[cpu] >> 4];
387                 show_leds(cpu);
388         }
389
390         profile_tick(CPU_PROFILING);
391
392         if(!--prof_counter(cpu)) {
393                 int user = user_mode(regs);
394
395                 irq_enter();
396                 update_process_times(user);
397                 irq_exit();
398
399                 prof_counter(cpu) = prof_multiplier(cpu);
400         }
401         set_irq_regs(old_regs);
402 }
403
404 extern unsigned int lvl14_resolution;
405
406 static void __init smp_setup_percpu_timer(void)
407 {
408         int cpu = hard_smp4d_processor_id();
409
410         prof_counter(cpu) = prof_multiplier(cpu) = 1;
411         load_profile_irq(cpu, lvl14_resolution);
412 }
413
414 void __init smp4d_blackbox_id(unsigned *addr)
415 {
416         int rd = *addr & 0x3e000000;
417         
418         addr[0] = 0xc0800800 | rd;              /* lda [%g0] ASI_M_VIKING_TMP1, reg */
419         addr[1] = 0x01000000;                   /* nop */
420         addr[2] = 0x01000000;                   /* nop */
421 }
422
423 void __init smp4d_blackbox_current(unsigned *addr)
424 {
425         int rd = *addr & 0x3e000000;
426         
427         addr[0] = 0xc0800800 | rd;              /* lda [%g0] ASI_M_VIKING_TMP1, reg */
428         addr[2] = 0x81282002 | rd | (rd >> 11); /* sll reg, 2, reg */
429         addr[4] = 0x01000000;                   /* nop */
430 }
431
432 void __init sun4d_init_smp(void)
433 {
434         int i;
435         extern unsigned int t_nmi[], linux_trap_ipi15_sun4d[], linux_trap_ipi15_sun4m[];
436
437         /* Patch ipi15 trap table */
438         t_nmi[1] = t_nmi[1] + (linux_trap_ipi15_sun4d - linux_trap_ipi15_sun4m);
439         
440         /* And set btfixup... */
441         BTFIXUPSET_BLACKBOX(hard_smp_processor_id, smp4d_blackbox_id);
442         BTFIXUPSET_BLACKBOX(load_current, smp4d_blackbox_current);
443         BTFIXUPSET_CALL(smp_cross_call, smp4d_cross_call, BTFIXUPCALL_NORM);
444         BTFIXUPSET_CALL(smp_message_pass, smp4d_message_pass, BTFIXUPCALL_NORM);
445         BTFIXUPSET_CALL(__hard_smp_processor_id, __smp4d_processor_id, BTFIXUPCALL_NORM);
446         
447         for (i = 0; i < NR_CPUS; i++) {
448                 ccall_info.processors_in[i] = 1;
449                 ccall_info.processors_out[i] = 1;
450         }
451 }