Merge commit 'gcl/gcl-next' into next
[linux-2.6] / arch / sparc / kernel / sun4m_smp.c
1 /* sun4m_smp.c: Sparc SUN4M SMP support.
2  *
3  * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu)
4  */
5
6 #include <asm/head.h>
7
8 #include <linux/kernel.h>
9 #include <linux/sched.h>
10 #include <linux/threads.h>
11 #include <linux/smp.h>
12 #include <linux/interrupt.h>
13 #include <linux/kernel_stat.h>
14 #include <linux/init.h>
15 #include <linux/spinlock.h>
16 #include <linux/mm.h>
17 #include <linux/swap.h>
18 #include <linux/profile.h>
19 #include <linux/delay.h>
20 #include <linux/cpu.h>
21
22 #include <asm/cacheflush.h>
23 #include <asm/tlbflush.h>
24 #include <asm/irq_regs.h>
25
26 #include <asm/ptrace.h>
27 #include <asm/atomic.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/cpudata.h>
35
36 #include "irq.h"
37
38 #define IRQ_CROSS_CALL          15
39
40 extern ctxd_t *srmmu_ctx_table_phys;
41
42 extern volatile unsigned long cpu_callin_map[NR_CPUS];
43 extern unsigned char boot_cpu_id;
44
45 extern cpumask_t smp_commenced_mask;
46
47 extern int __smp4m_processor_id(void);
48
49 /*#define SMP_DEBUG*/
50
51 #ifdef SMP_DEBUG
52 #define SMP_PRINTK(x)   printk x
53 #else
54 #define SMP_PRINTK(x)
55 #endif
56
57 static inline unsigned long
58 swap_ulong(volatile unsigned long *ptr, unsigned long val)
59 {
60         __asm__ __volatile__("swap [%1], %0\n\t" :
61                              "=&r" (val), "=&r" (ptr) :
62                              "0" (val), "1" (ptr));
63         return val;
64 }
65
66 static void smp_setup_percpu_timer(void);
67 extern void cpu_probe(void);
68
69 void __cpuinit smp4m_callin(void)
70 {
71         int cpuid = hard_smp_processor_id();
72
73         local_flush_cache_all();
74         local_flush_tlb_all();
75
76         notify_cpu_starting(cpuid);
77
78         /* Get our local ticker going. */
79         smp_setup_percpu_timer();
80
81         calibrate_delay();
82         smp_store_cpu_info(cpuid);
83
84         local_flush_cache_all();
85         local_flush_tlb_all();
86
87         /*
88          * Unblock the master CPU _only_ when the scheduler state
89          * of all secondary CPUs will be up-to-date, so after
90          * the SMP initialization the master will be just allowed
91          * to call the scheduler code.
92          */
93         /* Allow master to continue. */
94         swap_ulong(&cpu_callin_map[cpuid], 1);
95
96         /* XXX: What's up with all the flushes? */
97         local_flush_cache_all();
98         local_flush_tlb_all();
99         
100         cpu_probe();
101
102         /* Fix idle thread fields. */
103         __asm__ __volatile__("ld [%0], %%g6\n\t"
104                              : : "r" (&current_set[cpuid])
105                              : "memory" /* paranoid */);
106
107         /* Attach to the address space of init_task. */
108         atomic_inc(&init_mm.mm_count);
109         current->active_mm = &init_mm;
110
111         while (!cpu_isset(cpuid, smp_commenced_mask))
112                 mb();
113
114         local_irq_enable();
115
116         cpu_set(cpuid, cpu_online_map);
117 }
118
119 /*
120  *      Cycle through the processors asking the PROM to start each one.
121  */
122  
123 extern struct linux_prom_registers smp_penguin_ctable;
124 extern unsigned long trapbase_cpu1[];
125 extern unsigned long trapbase_cpu2[];
126 extern unsigned long trapbase_cpu3[];
127
128 void __init smp4m_boot_cpus(void)
129 {
130         smp_setup_percpu_timer();
131         local_flush_cache_all();
132 }
133
134 int __cpuinit smp4m_boot_one_cpu(int i)
135 {
136         extern unsigned long sun4m_cpu_startup;
137         unsigned long *entry = &sun4m_cpu_startup;
138         struct task_struct *p;
139         int timeout;
140         int cpu_node;
141
142         cpu_find_by_mid(i, &cpu_node);
143
144         /* Cook up an idler for this guy. */
145         p = fork_idle(i);
146         current_set[i] = task_thread_info(p);
147         /* See trampoline.S for details... */
148         entry += ((i-1) * 3);
149
150         /*
151          * Initialize the contexts table
152          * Since the call to prom_startcpu() trashes the structure,
153          * we need to re-initialize it for each cpu
154          */
155         smp_penguin_ctable.which_io = 0;
156         smp_penguin_ctable.phys_addr = (unsigned int) srmmu_ctx_table_phys;
157         smp_penguin_ctable.reg_size = 0;
158
159         /* whirrr, whirrr, whirrrrrrrrr... */
160         printk("Starting CPU %d at %p\n", i, entry);
161         local_flush_cache_all();
162         prom_startcpu(cpu_node,
163                       &smp_penguin_ctable, 0, (char *)entry);
164
165         /* wheee... it's going... */
166         for(timeout = 0; timeout < 10000; timeout++) {
167                 if(cpu_callin_map[i])
168                         break;
169                 udelay(200);
170         }
171
172         if (!(cpu_callin_map[i])) {
173                 printk("Processor %d is stuck.\n", i);
174                 return -ENODEV;
175         }
176
177         local_flush_cache_all();
178         return 0;
179 }
180
181 void __init smp4m_smp_done(void)
182 {
183         int i, first;
184         int *prev;
185
186         /* setup cpu list for irq rotation */
187         first = 0;
188         prev = &first;
189         for (i = 0; i < NR_CPUS; i++) {
190                 if (cpu_online(i)) {
191                         *prev = i;
192                         prev = &cpu_data(i).next;
193                 }
194         }
195         *prev = first;
196         local_flush_cache_all();
197
198         /* Free unneeded trap tables */
199         if (!cpu_isset(1, cpu_present_map)) {
200                 ClearPageReserved(virt_to_page(trapbase_cpu1));
201                 init_page_count(virt_to_page(trapbase_cpu1));
202                 free_page((unsigned long)trapbase_cpu1);
203                 totalram_pages++;
204                 num_physpages++;
205         }
206         if (!cpu_isset(2, cpu_present_map)) {
207                 ClearPageReserved(virt_to_page(trapbase_cpu2));
208                 init_page_count(virt_to_page(trapbase_cpu2));
209                 free_page((unsigned long)trapbase_cpu2);
210                 totalram_pages++;
211                 num_physpages++;
212         }
213         if (!cpu_isset(3, cpu_present_map)) {
214                 ClearPageReserved(virt_to_page(trapbase_cpu3));
215                 init_page_count(virt_to_page(trapbase_cpu3));
216                 free_page((unsigned long)trapbase_cpu3);
217                 totalram_pages++;
218                 num_physpages++;
219         }
220
221         /* Ok, they are spinning and ready to go. */
222 }
223
224 /* At each hardware IRQ, we get this called to forward IRQ reception
225  * to the next processor.  The caller must disable the IRQ level being
226  * serviced globally so that there are no double interrupts received.
227  *
228  * XXX See sparc64 irq.c.
229  */
230 void smp4m_irq_rotate(int cpu)
231 {
232         int next = cpu_data(cpu).next;
233         if (next != cpu)
234                 set_irq_udt(next);
235 }
236
237 static struct smp_funcall {
238         smpfunc_t func;
239         unsigned long arg1;
240         unsigned long arg2;
241         unsigned long arg3;
242         unsigned long arg4;
243         unsigned long arg5;
244         unsigned long processors_in[SUN4M_NCPUS];  /* Set when ipi entered. */
245         unsigned long processors_out[SUN4M_NCPUS]; /* Set when ipi exited. */
246 } ccall_info;
247
248 static DEFINE_SPINLOCK(cross_call_lock);
249
250 /* Cross calls must be serialized, at least currently. */
251 static void smp4m_cross_call(smpfunc_t func, cpumask_t mask, unsigned long arg1,
252                              unsigned long arg2, unsigned long arg3,
253                              unsigned long arg4)
254 {
255                 register int ncpus = SUN4M_NCPUS;
256                 unsigned long flags;
257
258                 spin_lock_irqsave(&cross_call_lock, flags);
259
260                 /* Init function glue. */
261                 ccall_info.func = func;
262                 ccall_info.arg1 = arg1;
263                 ccall_info.arg2 = arg2;
264                 ccall_info.arg3 = arg3;
265                 ccall_info.arg4 = arg4;
266                 ccall_info.arg5 = 0;
267
268                 /* Init receive/complete mapping, plus fire the IPI's off. */
269                 {
270                         register int i;
271
272                         cpu_clear(smp_processor_id(), mask);
273                         cpus_and(mask, cpu_online_map, mask);
274                         for(i = 0; i < ncpus; i++) {
275                                 if (cpu_isset(i, mask)) {
276                                         ccall_info.processors_in[i] = 0;
277                                         ccall_info.processors_out[i] = 0;
278                                         set_cpu_int(i, IRQ_CROSS_CALL);
279                                 } else {
280                                         ccall_info.processors_in[i] = 1;
281                                         ccall_info.processors_out[i] = 1;
282                                 }
283                         }
284                 }
285
286                 {
287                         register int i;
288
289                         i = 0;
290                         do {
291                                 if (!cpu_isset(i, mask))
292                                         continue;
293                                 while(!ccall_info.processors_in[i])
294                                         barrier();
295                         } while(++i < ncpus);
296
297                         i = 0;
298                         do {
299                                 if (!cpu_isset(i, mask))
300                                         continue;
301                                 while(!ccall_info.processors_out[i])
302                                         barrier();
303                         } while(++i < ncpus);
304                 }
305
306                 spin_unlock_irqrestore(&cross_call_lock, flags);
307 }
308
309 /* Running cross calls. */
310 void smp4m_cross_call_irq(void)
311 {
312         int i = smp_processor_id();
313
314         ccall_info.processors_in[i] = 1;
315         ccall_info.func(ccall_info.arg1, ccall_info.arg2, ccall_info.arg3,
316                         ccall_info.arg4, ccall_info.arg5);
317         ccall_info.processors_out[i] = 1;
318 }
319
320 extern void sun4m_clear_profile_irq(int cpu);
321
322 void smp4m_percpu_timer_interrupt(struct pt_regs *regs)
323 {
324         struct pt_regs *old_regs;
325         int cpu = smp_processor_id();
326
327         old_regs = set_irq_regs(regs);
328
329         sun4m_clear_profile_irq(cpu);
330
331         profile_tick(CPU_PROFILING);
332
333         if(!--prof_counter(cpu)) {
334                 int user = user_mode(regs);
335
336                 irq_enter();
337                 update_process_times(user);
338                 irq_exit();
339
340                 prof_counter(cpu) = prof_multiplier(cpu);
341         }
342         set_irq_regs(old_regs);
343 }
344
345 extern unsigned int lvl14_resolution;
346
347 static void __cpuinit smp_setup_percpu_timer(void)
348 {
349         int cpu = smp_processor_id();
350
351         prof_counter(cpu) = prof_multiplier(cpu) = 1;
352         load_profile_irq(cpu, lvl14_resolution);
353
354         if(cpu == boot_cpu_id)
355                 enable_pil_irq(14);
356 }
357
358 static void __init smp4m_blackbox_id(unsigned *addr)
359 {
360         int rd = *addr & 0x3e000000;
361         int rs1 = rd >> 11;
362         
363         addr[0] = 0x81580000 | rd;              /* rd %tbr, reg */
364         addr[1] = 0x8130200c | rd | rs1;        /* srl reg, 0xc, reg */
365         addr[2] = 0x80082003 | rd | rs1;        /* and reg, 3, reg */
366 }
367
368 static void __init smp4m_blackbox_current(unsigned *addr)
369 {
370         int rd = *addr & 0x3e000000;
371         int rs1 = rd >> 11;
372         
373         addr[0] = 0x81580000 | rd;              /* rd %tbr, reg */
374         addr[2] = 0x8130200a | rd | rs1;        /* srl reg, 0xa, reg */
375         addr[4] = 0x8008200c | rd | rs1;        /* and reg, 0xc, reg */
376 }
377
378 void __init sun4m_init_smp(void)
379 {
380         BTFIXUPSET_BLACKBOX(hard_smp_processor_id, smp4m_blackbox_id);
381         BTFIXUPSET_BLACKBOX(load_current, smp4m_blackbox_current);
382         BTFIXUPSET_CALL(smp_cross_call, smp4m_cross_call, BTFIXUPCALL_NORM);
383         BTFIXUPSET_CALL(__hard_smp_processor_id, __smp4m_processor_id, BTFIXUPCALL_NORM);
384 }