Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/arch/arm/kernel/smp.c | |
3 | * | |
4 | * Copyright (C) 2002 ARM Limited, All Rights Reserved. | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License version 2 as | |
8 | * published by the Free Software Foundation. | |
9 | */ | |
10 | #include <linux/config.h> | |
11 | #include <linux/delay.h> | |
12 | #include <linux/init.h> | |
13 | #include <linux/spinlock.h> | |
14 | #include <linux/sched.h> | |
15 | #include <linux/interrupt.h> | |
16 | #include <linux/cache.h> | |
17 | #include <linux/profile.h> | |
18 | #include <linux/errno.h> | |
19 | #include <linux/mm.h> | |
20 | #include <linux/cpu.h> | |
21 | #include <linux/smp.h> | |
22 | #include <linux/seq_file.h> | |
23 | ||
24 | #include <asm/atomic.h> | |
25 | #include <asm/cacheflush.h> | |
26 | #include <asm/cpu.h> | |
e65f38ed RK |
27 | #include <asm/mmu_context.h> |
28 | #include <asm/pgtable.h> | |
29 | #include <asm/pgalloc.h> | |
1da177e4 LT |
30 | #include <asm/processor.h> |
31 | #include <asm/tlbflush.h> | |
32 | #include <asm/ptrace.h> | |
33 | ||
34 | /* | |
35 | * bitmask of present and online CPUs. | |
36 | * The present bitmask indicates that the CPU is physically present. | |
37 | * The online bitmask indicates that the CPU is up and running. | |
38 | */ | |
d12734d1 | 39 | cpumask_t cpu_possible_map; |
1da177e4 LT |
40 | cpumask_t cpu_online_map; |
41 | ||
e65f38ed RK |
42 | /* |
43 | * as from 2.5, kernels no longer have an init_tasks structure | |
44 | * so we need some other way of telling a new secondary core | |
45 | * where to place its SVC stack | |
46 | */ | |
47 | struct secondary_data secondary_data; | |
48 | ||
1da177e4 LT |
49 | /* |
50 | * structures for inter-processor calls | |
51 | * - A collection of single bit ipi messages. | |
52 | */ | |
53 | struct ipi_data { | |
54 | spinlock_t lock; | |
55 | unsigned long ipi_count; | |
56 | unsigned long bits; | |
57 | }; | |
58 | ||
59 | static DEFINE_PER_CPU(struct ipi_data, ipi_data) = { | |
60 | .lock = SPIN_LOCK_UNLOCKED, | |
61 | }; | |
62 | ||
63 | enum ipi_msg_type { | |
64 | IPI_TIMER, | |
65 | IPI_RESCHEDULE, | |
66 | IPI_CALL_FUNC, | |
67 | IPI_CPU_STOP, | |
68 | }; | |
69 | ||
70 | struct smp_call_struct { | |
71 | void (*func)(void *info); | |
72 | void *info; | |
73 | int wait; | |
74 | cpumask_t pending; | |
75 | cpumask_t unfinished; | |
76 | }; | |
77 | ||
78 | static struct smp_call_struct * volatile smp_call_function_data; | |
79 | static DEFINE_SPINLOCK(smp_call_function_lock); | |
80 | ||
bd6f68af | 81 | int __cpuinit __cpu_up(unsigned int cpu) |
1da177e4 LT |
82 | { |
83 | struct task_struct *idle; | |
e65f38ed RK |
84 | pgd_t *pgd; |
85 | pmd_t *pmd; | |
1da177e4 LT |
86 | int ret; |
87 | ||
88 | /* | |
89 | * Spawn a new process manually. Grab a pointer to | |
90 | * its task struct so we can mess with it | |
91 | */ | |
92 | idle = fork_idle(cpu); | |
93 | if (IS_ERR(idle)) { | |
94 | printk(KERN_ERR "CPU%u: fork() failed\n", cpu); | |
95 | return PTR_ERR(idle); | |
96 | } | |
97 | ||
e65f38ed RK |
98 | /* |
99 | * Allocate initial page tables to allow the new CPU to | |
100 | * enable the MMU safely. This essentially means a set | |
101 | * of our "standard" page tables, with the addition of | |
102 | * a 1:1 mapping for the physical address of the kernel. | |
103 | */ | |
104 | pgd = pgd_alloc(&init_mm); | |
105 | pmd = pmd_offset(pgd, PHYS_OFFSET); | |
106 | *pmd = __pmd((PHYS_OFFSET & PGDIR_MASK) | | |
107 | PMD_TYPE_SECT | PMD_SECT_AP_WRITE); | |
108 | ||
109 | /* | |
110 | * We need to tell the secondary core where to find | |
111 | * its stack and the page tables. | |
112 | */ | |
113 | secondary_data.stack = (void *)idle->thread_info + THREAD_SIZE - 8; | |
114 | secondary_data.pgdir = virt_to_phys(pgd); | |
115 | wmb(); | |
116 | ||
1da177e4 LT |
117 | /* |
118 | * Now bring the CPU into our world. | |
119 | */ | |
120 | ret = boot_secondary(cpu, idle); | |
e65f38ed RK |
121 | if (ret == 0) { |
122 | unsigned long timeout; | |
123 | ||
124 | /* | |
125 | * CPU was successfully started, wait for it | |
126 | * to come online or time out. | |
127 | */ | |
128 | timeout = jiffies + HZ; | |
129 | while (time_before(jiffies, timeout)) { | |
130 | if (cpu_online(cpu)) | |
131 | break; | |
132 | ||
133 | udelay(10); | |
134 | barrier(); | |
135 | } | |
136 | ||
137 | if (!cpu_online(cpu)) | |
138 | ret = -EIO; | |
139 | } | |
140 | ||
141 | secondary_data.stack = 0; | |
142 | secondary_data.pgdir = 0; | |
143 | ||
144 | *pmd_offset(pgd, PHYS_OFFSET) = __pmd(0); | |
145 | pgd_free(pgd); | |
146 | ||
1da177e4 | 147 | if (ret) { |
0908db22 RK |
148 | printk(KERN_CRIT "CPU%u: processor failed to boot\n", cpu); |
149 | ||
1da177e4 LT |
150 | /* |
151 | * FIXME: We need to clean up the new idle thread. --rmk | |
152 | */ | |
153 | } | |
154 | ||
155 | return ret; | |
156 | } | |
157 | ||
e65f38ed RK |
158 | /* |
159 | * This is the secondary CPU boot entry. We're using this CPUs | |
160 | * idle thread stack, but a set of temporary page tables. | |
161 | */ | |
bd6f68af | 162 | asmlinkage void __cpuinit secondary_start_kernel(void) |
e65f38ed RK |
163 | { |
164 | struct mm_struct *mm = &init_mm; | |
165 | unsigned int cpu = smp_processor_id(); | |
166 | ||
167 | printk("CPU%u: Booted secondary processor\n", cpu); | |
168 | ||
169 | /* | |
170 | * All kernel threads share the same mm context; grab a | |
171 | * reference and switch to it. | |
172 | */ | |
173 | atomic_inc(&mm->mm_users); | |
174 | atomic_inc(&mm->mm_count); | |
175 | current->active_mm = mm; | |
176 | cpu_set(cpu, mm->cpu_vm_mask); | |
177 | cpu_switch_mm(mm->pgd, mm); | |
178 | enter_lazy_tlb(mm, current); | |
505d7b19 | 179 | local_flush_tlb_all(); |
e65f38ed RK |
180 | |
181 | cpu_init(); | |
182 | ||
183 | /* | |
184 | * Give the platform a chance to do its own initialisation. | |
185 | */ | |
186 | platform_secondary_init(cpu); | |
187 | ||
188 | /* | |
189 | * Enable local interrupts. | |
190 | */ | |
191 | local_irq_enable(); | |
192 | local_fiq_enable(); | |
193 | ||
194 | calibrate_delay(); | |
195 | ||
196 | smp_store_cpu_info(cpu); | |
197 | ||
198 | /* | |
199 | * OK, now it's safe to let the boot CPU continue | |
200 | */ | |
201 | cpu_set(cpu, cpu_online_map); | |
202 | ||
203 | /* | |
204 | * OK, it's off to the idle thread for us | |
205 | */ | |
206 | cpu_idle(); | |
207 | } | |
208 | ||
1da177e4 LT |
209 | /* |
210 | * Called by both boot and secondaries to move global data into | |
211 | * per-processor storage. | |
212 | */ | |
bd6f68af | 213 | void __cpuinit smp_store_cpu_info(unsigned int cpuid) |
1da177e4 LT |
214 | { |
215 | struct cpuinfo_arm *cpu_info = &per_cpu(cpu_data, cpuid); | |
216 | ||
217 | cpu_info->loops_per_jiffy = loops_per_jiffy; | |
218 | } | |
219 | ||
220 | void __init smp_cpus_done(unsigned int max_cpus) | |
221 | { | |
222 | int cpu; | |
223 | unsigned long bogosum = 0; | |
224 | ||
225 | for_each_online_cpu(cpu) | |
226 | bogosum += per_cpu(cpu_data, cpu).loops_per_jiffy; | |
227 | ||
228 | printk(KERN_INFO "SMP: Total of %d processors activated " | |
229 | "(%lu.%02lu BogoMIPS).\n", | |
230 | num_online_cpus(), | |
231 | bogosum / (500000/HZ), | |
232 | (bogosum / (5000/HZ)) % 100); | |
233 | } | |
234 | ||
235 | void __init smp_prepare_boot_cpu(void) | |
236 | { | |
237 | unsigned int cpu = smp_processor_id(); | |
238 | ||
d12734d1 | 239 | cpu_set(cpu, cpu_possible_map); |
73eb7d9e | 240 | cpu_set(cpu, cpu_present_map); |
1da177e4 LT |
241 | cpu_set(cpu, cpu_online_map); |
242 | } | |
243 | ||
244 | static void send_ipi_message(cpumask_t callmap, enum ipi_msg_type msg) | |
245 | { | |
246 | unsigned long flags; | |
247 | unsigned int cpu; | |
248 | ||
249 | local_irq_save(flags); | |
250 | ||
251 | for_each_cpu_mask(cpu, callmap) { | |
252 | struct ipi_data *ipi = &per_cpu(ipi_data, cpu); | |
253 | ||
254 | spin_lock(&ipi->lock); | |
255 | ipi->bits |= 1 << msg; | |
256 | spin_unlock(&ipi->lock); | |
257 | } | |
258 | ||
259 | /* | |
260 | * Call the platform specific cross-CPU call function. | |
261 | */ | |
262 | smp_cross_call(callmap); | |
263 | ||
264 | local_irq_restore(flags); | |
265 | } | |
266 | ||
267 | /* | |
268 | * You must not call this function with disabled interrupts, from a | |
269 | * hardware interrupt handler, nor from a bottom half handler. | |
270 | */ | |
271 | int smp_call_function_on_cpu(void (*func)(void *info), void *info, int retry, | |
272 | int wait, cpumask_t callmap) | |
273 | { | |
274 | struct smp_call_struct data; | |
275 | unsigned long timeout; | |
276 | int ret = 0; | |
277 | ||
278 | data.func = func; | |
279 | data.info = info; | |
280 | data.wait = wait; | |
281 | ||
282 | cpu_clear(smp_processor_id(), callmap); | |
283 | if (cpus_empty(callmap)) | |
284 | goto out; | |
285 | ||
286 | data.pending = callmap; | |
287 | if (wait) | |
288 | data.unfinished = callmap; | |
289 | ||
290 | /* | |
291 | * try to get the mutex on smp_call_function_data | |
292 | */ | |
293 | spin_lock(&smp_call_function_lock); | |
294 | smp_call_function_data = &data; | |
295 | ||
296 | send_ipi_message(callmap, IPI_CALL_FUNC); | |
297 | ||
298 | timeout = jiffies + HZ; | |
299 | while (!cpus_empty(data.pending) && time_before(jiffies, timeout)) | |
300 | barrier(); | |
301 | ||
302 | /* | |
303 | * did we time out? | |
304 | */ | |
305 | if (!cpus_empty(data.pending)) { | |
306 | /* | |
307 | * this may be causing our panic - report it | |
308 | */ | |
309 | printk(KERN_CRIT | |
310 | "CPU%u: smp_call_function timeout for %p(%p)\n" | |
311 | " callmap %lx pending %lx, %swait\n", | |
312 | smp_processor_id(), func, info, callmap, data.pending, | |
313 | wait ? "" : "no "); | |
314 | ||
315 | /* | |
316 | * TRACE | |
317 | */ | |
318 | timeout = jiffies + (5 * HZ); | |
319 | while (!cpus_empty(data.pending) && time_before(jiffies, timeout)) | |
320 | barrier(); | |
321 | ||
322 | if (cpus_empty(data.pending)) | |
323 | printk(KERN_CRIT " RESOLVED\n"); | |
324 | else | |
325 | printk(KERN_CRIT " STILL STUCK\n"); | |
326 | } | |
327 | ||
328 | /* | |
329 | * whatever happened, we're done with the data, so release it | |
330 | */ | |
331 | smp_call_function_data = NULL; | |
332 | spin_unlock(&smp_call_function_lock); | |
333 | ||
334 | if (!cpus_empty(data.pending)) { | |
335 | ret = -ETIMEDOUT; | |
336 | goto out; | |
337 | } | |
338 | ||
339 | if (wait) | |
340 | while (!cpus_empty(data.unfinished)) | |
341 | barrier(); | |
342 | out: | |
343 | ||
344 | return 0; | |
345 | } | |
346 | ||
347 | int smp_call_function(void (*func)(void *info), void *info, int retry, | |
348 | int wait) | |
349 | { | |
350 | return smp_call_function_on_cpu(func, info, retry, wait, | |
351 | cpu_online_map); | |
352 | } | |
353 | ||
354 | void show_ipi_list(struct seq_file *p) | |
355 | { | |
356 | unsigned int cpu; | |
357 | ||
358 | seq_puts(p, "IPI:"); | |
359 | ||
e11b2236 | 360 | for_each_present_cpu(cpu) |
1da177e4 LT |
361 | seq_printf(p, " %10lu", per_cpu(ipi_data, cpu).ipi_count); |
362 | ||
363 | seq_putc(p, '\n'); | |
364 | } | |
365 | ||
366 | static void ipi_timer(struct pt_regs *regs) | |
367 | { | |
368 | int user = user_mode(regs); | |
369 | ||
370 | irq_enter(); | |
371 | profile_tick(CPU_PROFILING, regs); | |
372 | update_process_times(user); | |
373 | irq_exit(); | |
374 | } | |
375 | ||
376 | /* | |
377 | * ipi_call_function - handle IPI from smp_call_function() | |
378 | * | |
379 | * Note that we copy data out of the cross-call structure and then | |
380 | * let the caller know that we're here and have done with their data | |
381 | */ | |
382 | static void ipi_call_function(unsigned int cpu) | |
383 | { | |
384 | struct smp_call_struct *data = smp_call_function_data; | |
385 | void (*func)(void *info) = data->func; | |
386 | void *info = data->info; | |
387 | int wait = data->wait; | |
388 | ||
389 | cpu_clear(cpu, data->pending); | |
390 | ||
391 | func(info); | |
392 | ||
393 | if (wait) | |
394 | cpu_clear(cpu, data->unfinished); | |
395 | } | |
396 | ||
397 | static DEFINE_SPINLOCK(stop_lock); | |
398 | ||
399 | /* | |
400 | * ipi_cpu_stop - handle IPI from smp_send_stop() | |
401 | */ | |
402 | static void ipi_cpu_stop(unsigned int cpu) | |
403 | { | |
404 | spin_lock(&stop_lock); | |
405 | printk(KERN_CRIT "CPU%u: stopping\n", cpu); | |
406 | dump_stack(); | |
407 | spin_unlock(&stop_lock); | |
408 | ||
409 | cpu_clear(cpu, cpu_online_map); | |
410 | ||
411 | local_fiq_disable(); | |
412 | local_irq_disable(); | |
413 | ||
414 | while (1) | |
415 | cpu_relax(); | |
416 | } | |
417 | ||
418 | /* | |
419 | * Main handler for inter-processor interrupts | |
420 | * | |
421 | * For ARM, the ipimask now only identifies a single | |
422 | * category of IPI (Bit 1 IPIs have been replaced by a | |
423 | * different mechanism): | |
424 | * | |
425 | * Bit 0 - Inter-processor function call | |
426 | */ | |
427 | void do_IPI(struct pt_regs *regs) | |
428 | { | |
429 | unsigned int cpu = smp_processor_id(); | |
430 | struct ipi_data *ipi = &per_cpu(ipi_data, cpu); | |
431 | ||
432 | ipi->ipi_count++; | |
433 | ||
434 | for (;;) { | |
435 | unsigned long msgs; | |
436 | ||
437 | spin_lock(&ipi->lock); | |
438 | msgs = ipi->bits; | |
439 | ipi->bits = 0; | |
440 | spin_unlock(&ipi->lock); | |
441 | ||
442 | if (!msgs) | |
443 | break; | |
444 | ||
445 | do { | |
446 | unsigned nextmsg; | |
447 | ||
448 | nextmsg = msgs & -msgs; | |
449 | msgs &= ~nextmsg; | |
450 | nextmsg = ffz(~nextmsg); | |
451 | ||
452 | switch (nextmsg) { | |
453 | case IPI_TIMER: | |
454 | ipi_timer(regs); | |
455 | break; | |
456 | ||
457 | case IPI_RESCHEDULE: | |
458 | /* | |
459 | * nothing more to do - eveything is | |
460 | * done on the interrupt return path | |
461 | */ | |
462 | break; | |
463 | ||
464 | case IPI_CALL_FUNC: | |
465 | ipi_call_function(cpu); | |
466 | break; | |
467 | ||
468 | case IPI_CPU_STOP: | |
469 | ipi_cpu_stop(cpu); | |
470 | break; | |
471 | ||
472 | default: | |
473 | printk(KERN_CRIT "CPU%u: Unknown IPI message 0x%x\n", | |
474 | cpu, nextmsg); | |
475 | break; | |
476 | } | |
477 | } while (msgs); | |
478 | } | |
479 | } | |
480 | ||
481 | void smp_send_reschedule(int cpu) | |
482 | { | |
483 | send_ipi_message(cpumask_of_cpu(cpu), IPI_RESCHEDULE); | |
484 | } | |
485 | ||
486 | void smp_send_timer(void) | |
487 | { | |
488 | cpumask_t mask = cpu_online_map; | |
489 | cpu_clear(smp_processor_id(), mask); | |
490 | send_ipi_message(mask, IPI_TIMER); | |
491 | } | |
492 | ||
493 | void smp_send_stop(void) | |
494 | { | |
495 | cpumask_t mask = cpu_online_map; | |
496 | cpu_clear(smp_processor_id(), mask); | |
497 | send_ipi_message(mask, IPI_CPU_STOP); | |
498 | } | |
499 | ||
500 | /* | |
501 | * not supported here | |
502 | */ | |
503 | int __init setup_profiling_timer(unsigned int multiplier) | |
504 | { | |
505 | return -EINVAL; | |
506 | } | |
4b0ef3b1 RK |
507 | |
508 | static int | |
509 | on_each_cpu_mask(void (*func)(void *), void *info, int retry, int wait, | |
510 | cpumask_t mask) | |
511 | { | |
512 | int ret = 0; | |
513 | ||
514 | preempt_disable(); | |
515 | ||
516 | ret = smp_call_function_on_cpu(func, info, retry, wait, mask); | |
517 | if (cpu_isset(smp_processor_id(), mask)) | |
518 | func(info); | |
519 | ||
520 | preempt_enable(); | |
521 | ||
522 | return ret; | |
523 | } | |
524 | ||
525 | /**********************************************************************/ | |
526 | ||
527 | /* | |
528 | * TLB operations | |
529 | */ | |
530 | struct tlb_args { | |
531 | struct vm_area_struct *ta_vma; | |
532 | unsigned long ta_start; | |
533 | unsigned long ta_end; | |
534 | }; | |
535 | ||
536 | static inline void ipi_flush_tlb_all(void *ignored) | |
537 | { | |
538 | local_flush_tlb_all(); | |
539 | } | |
540 | ||
541 | static inline void ipi_flush_tlb_mm(void *arg) | |
542 | { | |
543 | struct mm_struct *mm = (struct mm_struct *)arg; | |
544 | ||
545 | local_flush_tlb_mm(mm); | |
546 | } | |
547 | ||
548 | static inline void ipi_flush_tlb_page(void *arg) | |
549 | { | |
550 | struct tlb_args *ta = (struct tlb_args *)arg; | |
551 | ||
552 | local_flush_tlb_page(ta->ta_vma, ta->ta_start); | |
553 | } | |
554 | ||
555 | static inline void ipi_flush_tlb_kernel_page(void *arg) | |
556 | { | |
557 | struct tlb_args *ta = (struct tlb_args *)arg; | |
558 | ||
559 | local_flush_tlb_kernel_page(ta->ta_start); | |
560 | } | |
561 | ||
562 | static inline void ipi_flush_tlb_range(void *arg) | |
563 | { | |
564 | struct tlb_args *ta = (struct tlb_args *)arg; | |
565 | ||
566 | local_flush_tlb_range(ta->ta_vma, ta->ta_start, ta->ta_end); | |
567 | } | |
568 | ||
569 | static inline void ipi_flush_tlb_kernel_range(void *arg) | |
570 | { | |
571 | struct tlb_args *ta = (struct tlb_args *)arg; | |
572 | ||
573 | local_flush_tlb_kernel_range(ta->ta_start, ta->ta_end); | |
574 | } | |
575 | ||
576 | void flush_tlb_all(void) | |
577 | { | |
578 | on_each_cpu(ipi_flush_tlb_all, NULL, 1, 1); | |
579 | } | |
580 | ||
581 | void flush_tlb_mm(struct mm_struct *mm) | |
582 | { | |
583 | cpumask_t mask = mm->cpu_vm_mask; | |
584 | ||
585 | on_each_cpu_mask(ipi_flush_tlb_mm, mm, 1, 1, mask); | |
586 | } | |
587 | ||
588 | void flush_tlb_page(struct vm_area_struct *vma, unsigned long uaddr) | |
589 | { | |
590 | cpumask_t mask = vma->vm_mm->cpu_vm_mask; | |
591 | struct tlb_args ta; | |
592 | ||
593 | ta.ta_vma = vma; | |
594 | ta.ta_start = uaddr; | |
595 | ||
596 | on_each_cpu_mask(ipi_flush_tlb_page, &ta, 1, 1, mask); | |
597 | } | |
598 | ||
599 | void flush_tlb_kernel_page(unsigned long kaddr) | |
600 | { | |
601 | struct tlb_args ta; | |
602 | ||
603 | ta.ta_start = kaddr; | |
604 | ||
605 | on_each_cpu(ipi_flush_tlb_kernel_page, &ta, 1, 1); | |
606 | } | |
607 | ||
608 | void flush_tlb_range(struct vm_area_struct *vma, | |
609 | unsigned long start, unsigned long end) | |
610 | { | |
611 | cpumask_t mask = vma->vm_mm->cpu_vm_mask; | |
612 | struct tlb_args ta; | |
613 | ||
614 | ta.ta_vma = vma; | |
615 | ta.ta_start = start; | |
616 | ta.ta_end = end; | |
617 | ||
618 | on_each_cpu_mask(ipi_flush_tlb_range, &ta, 1, 1, mask); | |
619 | } | |
620 | ||
621 | void flush_tlb_kernel_range(unsigned long start, unsigned long end) | |
622 | { | |
623 | struct tlb_args ta; | |
624 | ||
625 | ta.ta_start = start; | |
626 | ta.ta_end = end; | |
627 | ||
628 | on_each_cpu(ipi_flush_tlb_kernel_range, &ta, 1, 1); | |
629 | } |