Merge branch 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/djbw/async_tx
[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 #include <linux/delay.h>
23 #include <linux/cpu.h>
24
25 #include <asm/ptrace.h>
26 #include <asm/atomic.h>
27 #include <asm/irq_regs.h>
28
29 #include <asm/irq.h>
30 #include <asm/page.h>
31 #include <asm/pgalloc.h>
32 #include <asm/pgtable.h>
33 #include <asm/oplib.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 static volatile int smp_processors_ready = 0;
45 static int smp_highest_cpu;
46 extern volatile unsigned long cpu_callin_map[NR_CPUS];
47 extern cpuinfo_sparc cpu_data[NR_CPUS];
48 extern unsigned char boot_cpu_id;
49 extern volatile int smp_process_available;
50
51 extern cpumask_t smp_commenced_mask;
52
53 extern int __smp4d_processor_id(void);
54
55 /* #define SMP_DEBUG */
56
57 #ifdef SMP_DEBUG
58 #define SMP_PRINTK(x)   printk x
59 #else
60 #define SMP_PRINTK(x)
61 #endif
62
63 static inline unsigned long swap(volatile unsigned long *ptr, unsigned long val)
64 {
65         __asm__ __volatile__("swap [%1], %0\n\t" :
66                              "=&r" (val), "=&r" (ptr) :
67                              "0" (val), "1" (ptr));
68         return val;
69 }
70
71 static void smp_setup_percpu_timer(void);
72 extern void cpu_probe(void);
73 extern void sun4d_distribute_irqs(void);
74
75 static unsigned char cpu_leds[32];
76
77 static inline void show_leds(int cpuid)
78 {
79         cpuid &= 0x1e;
80         __asm__ __volatile__ ("stba %0, [%1] %2" : :
81                               "r" ((cpu_leds[cpuid] << 4) | cpu_leds[cpuid+1]),
82                               "r" (ECSR_BASE(cpuid) | BB_LEDS),
83                               "i" (ASI_M_CTL));
84 }
85
86 void __init smp4d_callin(void)
87 {
88         int cpuid = hard_smp4d_processor_id();
89         extern spinlock_t sun4d_imsk_lock;
90         unsigned long flags;
91         
92         /* Show we are alive */
93         cpu_leds[cpuid] = 0x6;
94         show_leds(cpuid);
95
96         /* Enable level15 interrupt, disable level14 interrupt for now */
97         cc_set_imsk((cc_get_imsk() & ~0x8000) | 0x4000);
98
99         local_flush_cache_all();
100         local_flush_tlb_all();
101
102         notify_cpu_starting(cpuid);
103         /*
104          * Unblock the master CPU _only_ when the scheduler state
105          * of all secondary CPUs will be up-to-date, so after
106          * the SMP initialization the master will be just allowed
107          * to call the scheduler code.
108          */
109         /* Get our local ticker going. */
110         smp_setup_percpu_timer();
111
112         calibrate_delay();
113         smp_store_cpu_info(cpuid);
114         local_flush_cache_all();
115         local_flush_tlb_all();
116
117         /* Allow master to continue. */
118         swap((unsigned long *)&cpu_callin_map[cpuid], 1);
119         local_flush_cache_all();
120         local_flush_tlb_all();
121         
122         cpu_probe();
123
124         while((unsigned long)current_set[cpuid] < PAGE_OFFSET)
125                 barrier();
126                 
127         while(current_set[cpuid]->cpu != cpuid)
128                 barrier();
129                 
130         /* Fix idle thread fields. */
131         __asm__ __volatile__("ld [%0], %%g6\n\t"
132                              : : "r" (&current_set[cpuid])
133                              : "memory" /* paranoid */);
134
135         cpu_leds[cpuid] = 0x9;
136         show_leds(cpuid);
137         
138         /* Attach to the address space of init_task. */
139         atomic_inc(&init_mm.mm_count);
140         current->active_mm = &init_mm;
141
142         local_flush_cache_all();
143         local_flush_tlb_all();
144         
145         local_irq_enable();     /* We don't allow PIL 14 yet */
146         
147         while (!cpu_isset(cpuid, smp_commenced_mask))
148                 barrier();
149
150         spin_lock_irqsave(&sun4d_imsk_lock, flags);
151         cc_set_imsk(cc_get_imsk() & ~0x4000); /* Allow PIL 14 as well */
152         spin_unlock_irqrestore(&sun4d_imsk_lock, flags);
153         cpu_set(cpuid, cpu_online_map);
154
155 }
156
157 extern void init_IRQ(void);
158 extern void cpu_panic(void);
159
160 /*
161  *      Cycle through the processors asking the PROM to start each one.
162  */
163  
164 extern struct linux_prom_registers smp_penguin_ctable;
165 extern unsigned long trapbase_cpu1[];
166 extern unsigned long trapbase_cpu2[];
167 extern unsigned long trapbase_cpu3[];
168
169 void __init smp4d_boot_cpus(void)
170 {
171         if (boot_cpu_id)
172                 current_set[0] = NULL;
173         smp_setup_percpu_timer();
174         local_flush_cache_all();
175 }
176
177 int __cpuinit smp4d_boot_one_cpu(int i)
178 {
179                         extern unsigned long sun4d_cpu_startup;
180                         unsigned long *entry = &sun4d_cpu_startup;
181                         struct task_struct *p;
182                         int timeout;
183                         int cpu_node;
184
185                         cpu_find_by_instance(i, &cpu_node,NULL);
186                         /* Cook up an idler for this guy. */
187                         p = fork_idle(i);
188                         current_set[i] = task_thread_info(p);
189
190                         /*
191                          * Initialize the contexts table
192                          * Since the call to prom_startcpu() trashes the structure,
193                          * we need to re-initialize it for each cpu
194                          */
195                         smp_penguin_ctable.which_io = 0;
196                         smp_penguin_ctable.phys_addr = (unsigned int) srmmu_ctx_table_phys;
197                         smp_penguin_ctable.reg_size = 0;
198
199                         /* whirrr, whirrr, whirrrrrrrrr... */
200                         SMP_PRINTK(("Starting CPU %d at %p \n", i, entry));
201                         local_flush_cache_all();
202                         prom_startcpu(cpu_node,
203                                       &smp_penguin_ctable, 0, (char *)entry);
204                                       
205                         SMP_PRINTK(("prom_startcpu returned :)\n"));
206
207                         /* wheee... it's going... */
208                         for(timeout = 0; timeout < 10000; timeout++) {
209                                 if(cpu_callin_map[i])
210                                         break;
211                                 udelay(200);
212                         }
213                         
214         if (!(cpu_callin_map[i])) {
215                 printk("Processor %d is stuck.\n", i);
216                 return -ENODEV;
217
218         }
219         local_flush_cache_all();
220         return 0;
221 }
222
223 void __init smp4d_smp_done(void)
224 {
225         int i, first;
226         int *prev;
227
228         /* setup cpu list for irq rotation */
229         first = 0;
230         prev = &first;
231         for (i = 0; i < NR_CPUS; i++)
232                 if (cpu_online(i)) {
233                         *prev = i;
234                         prev = &cpu_data(i).next;
235                 }
236         *prev = first;
237         local_flush_cache_all();
238
239         /* Free unneeded trap tables */
240         ClearPageReserved(virt_to_page(trapbase_cpu1));
241         init_page_count(virt_to_page(trapbase_cpu1));
242         free_page((unsigned long)trapbase_cpu1);
243         totalram_pages++;
244         num_physpages++;
245
246         ClearPageReserved(virt_to_page(trapbase_cpu2));
247         init_page_count(virt_to_page(trapbase_cpu2));
248         free_page((unsigned long)trapbase_cpu2);
249         totalram_pages++;
250         num_physpages++;
251
252         ClearPageReserved(virt_to_page(trapbase_cpu3));
253         init_page_count(virt_to_page(trapbase_cpu3));
254         free_page((unsigned long)trapbase_cpu3);
255         totalram_pages++;
256         num_physpages++;
257
258         /* Ok, they are spinning and ready to go. */
259         smp_processors_ready = 1;
260         sun4d_distribute_irqs();
261 }
262
263 static struct smp_funcall {
264         smpfunc_t func;
265         unsigned long arg1;
266         unsigned long arg2;
267         unsigned long arg3;
268         unsigned long arg4;
269         unsigned long arg5;
270         unsigned char processors_in[NR_CPUS];  /* Set when ipi entered. */
271         unsigned char processors_out[NR_CPUS]; /* Set when ipi exited. */
272 } ccall_info __attribute__((aligned(8)));
273
274 static DEFINE_SPINLOCK(cross_call_lock);
275
276 /* Cross calls must be serialized, at least currently. */
277 static void smp4d_cross_call(smpfunc_t func, cpumask_t mask, unsigned long arg1,
278                              unsigned long arg2, unsigned long arg3,
279                              unsigned long arg4)
280 {
281         if(smp_processors_ready) {
282                 register int high = smp_highest_cpu;
283                 unsigned long flags;
284
285                 spin_lock_irqsave(&cross_call_lock, flags);
286
287                 {
288                         /* If you make changes here, make sure gcc generates proper code... */
289                         register smpfunc_t f asm("i0") = func;
290                         register unsigned long a1 asm("i1") = arg1;
291                         register unsigned long a2 asm("i2") = arg2;
292                         register unsigned long a3 asm("i3") = arg3;
293                         register unsigned long a4 asm("i4") = arg4;
294                         register unsigned long a5 asm("i5") = 0;
295
296                         __asm__ __volatile__(
297                                 "std %0, [%6]\n\t"
298                                 "std %2, [%6 + 8]\n\t"
299                                 "std %4, [%6 + 16]\n\t" : :
300                                 "r"(f), "r"(a1), "r"(a2), "r"(a3), "r"(a4), "r"(a5),
301                                 "r" (&ccall_info.func));
302                 }
303
304                 /* Init receive/complete mapping, plus fire the IPI's off. */
305                 {
306                         register int i;
307
308                         cpu_clear(smp_processor_id(), mask);
309                         cpus_and(mask, cpu_online_map, mask);
310                         for(i = 0; i <= high; i++) {
311                                 if (cpu_isset(i, mask)) {
312                                         ccall_info.processors_in[i] = 0;
313                                         ccall_info.processors_out[i] = 0;
314                                         sun4d_send_ipi(i, IRQ_CROSS_CALL);
315                                 }
316                         }
317                 }
318
319                 {
320                         register int i;
321
322                         i = 0;
323                         do {
324                                 if (!cpu_isset(i, mask))
325                                         continue;
326                                 while(!ccall_info.processors_in[i])
327                                         barrier();
328                         } while(++i <= high);
329
330                         i = 0;
331                         do {
332                                 if (!cpu_isset(i, mask))
333                                         continue;
334                                 while(!ccall_info.processors_out[i])
335                                         barrier();
336                         } while(++i <= high);
337                 }
338
339                 spin_unlock_irqrestore(&cross_call_lock, flags);
340         }
341 }
342
343 /* Running cross calls. */
344 void smp4d_cross_call_irq(void)
345 {
346         int i = hard_smp4d_processor_id();
347
348         ccall_info.processors_in[i] = 1;
349         ccall_info.func(ccall_info.arg1, ccall_info.arg2, ccall_info.arg3,
350                         ccall_info.arg4, ccall_info.arg5);
351         ccall_info.processors_out[i] = 1;
352 }
353
354 void smp4d_percpu_timer_interrupt(struct pt_regs *regs)
355 {
356         struct pt_regs *old_regs;
357         int cpu = hard_smp4d_processor_id();
358         static int cpu_tick[NR_CPUS];
359         static char led_mask[] = { 0xe, 0xd, 0xb, 0x7, 0xb, 0xd };
360
361         old_regs = set_irq_regs(regs);
362         bw_get_prof_limit(cpu); 
363         bw_clear_intr_mask(0, 1);       /* INTR_TABLE[0] & 1 is Profile IRQ */
364
365         cpu_tick[cpu]++;
366         if (!(cpu_tick[cpu] & 15)) {
367                 if (cpu_tick[cpu] == 0x60)
368                         cpu_tick[cpu] = 0;
369                 cpu_leds[cpu] = led_mask[cpu_tick[cpu] >> 4];
370                 show_leds(cpu);
371         }
372
373         profile_tick(CPU_PROFILING);
374
375         if(!--prof_counter(cpu)) {
376                 int user = user_mode(regs);
377
378                 irq_enter();
379                 update_process_times(user);
380                 irq_exit();
381
382                 prof_counter(cpu) = prof_multiplier(cpu);
383         }
384         set_irq_regs(old_regs);
385 }
386
387 extern unsigned int lvl14_resolution;
388
389 static void __init smp_setup_percpu_timer(void)
390 {
391         int cpu = hard_smp4d_processor_id();
392
393         prof_counter(cpu) = prof_multiplier(cpu) = 1;
394         load_profile_irq(cpu, lvl14_resolution);
395 }
396
397 void __init smp4d_blackbox_id(unsigned *addr)
398 {
399         int rd = *addr & 0x3e000000;
400         
401         addr[0] = 0xc0800800 | rd;              /* lda [%g0] ASI_M_VIKING_TMP1, reg */
402         addr[1] = 0x01000000;                   /* nop */
403         addr[2] = 0x01000000;                   /* nop */
404 }
405
406 void __init smp4d_blackbox_current(unsigned *addr)
407 {
408         int rd = *addr & 0x3e000000;
409         
410         addr[0] = 0xc0800800 | rd;              /* lda [%g0] ASI_M_VIKING_TMP1, reg */
411         addr[2] = 0x81282002 | rd | (rd >> 11); /* sll reg, 2, reg */
412         addr[4] = 0x01000000;                   /* nop */
413 }
414
415 void __init sun4d_init_smp(void)
416 {
417         int i;
418         extern unsigned int t_nmi[], linux_trap_ipi15_sun4d[], linux_trap_ipi15_sun4m[];
419
420         /* Patch ipi15 trap table */
421         t_nmi[1] = t_nmi[1] + (linux_trap_ipi15_sun4d - linux_trap_ipi15_sun4m);
422         
423         /* And set btfixup... */
424         BTFIXUPSET_BLACKBOX(hard_smp_processor_id, smp4d_blackbox_id);
425         BTFIXUPSET_BLACKBOX(load_current, smp4d_blackbox_current);
426         BTFIXUPSET_CALL(smp_cross_call, smp4d_cross_call, BTFIXUPCALL_NORM);
427         BTFIXUPSET_CALL(__hard_smp_processor_id, __smp4d_processor_id, BTFIXUPCALL_NORM);
428         
429         for (i = 0; i < NR_CPUS; i++) {
430                 ccall_info.processors_in[i] = 1;
431                 ccall_info.processors_out[i] = 1;
432         }
433 }