diff options
Diffstat (limited to 'arch/arm64/kernel/smp.c')
-rw-r--r-- | arch/arm64/kernel/smp.c | 469 |
1 files changed, 469 insertions, 0 deletions
diff --git a/arch/arm64/kernel/smp.c b/arch/arm64/kernel/smp.c new file mode 100644 index 000000000000..b711525be21f --- /dev/null +++ b/arch/arm64/kernel/smp.c | |||
@@ -0,0 +1,469 @@ | |||
1 | /* | ||
2 | * SMP initialisation and IPI support | ||
3 | * Based on arch/arm/kernel/smp.c | ||
4 | * | ||
5 | * Copyright (C) 2012 ARM Ltd. | ||
6 | * | ||
7 | * This program is free software; you can redistribute it and/or modify | ||
8 | * it under the terms of the GNU General Public License version 2 as | ||
9 | * published by the Free Software Foundation. | ||
10 | * | ||
11 | * This program is distributed in the hope that it will be useful, | ||
12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
14 | * GNU General Public License for more details. | ||
15 | * | ||
16 | * You should have received a copy of the GNU General Public License | ||
17 | * along with this program. If not, see <http://www.gnu.org/licenses/>. | ||
18 | */ | ||
19 | |||
20 | #include <linux/delay.h> | ||
21 | #include <linux/init.h> | ||
22 | #include <linux/spinlock.h> | ||
23 | #include <linux/sched.h> | ||
24 | #include <linux/interrupt.h> | ||
25 | #include <linux/cache.h> | ||
26 | #include <linux/profile.h> | ||
27 | #include <linux/errno.h> | ||
28 | #include <linux/mm.h> | ||
29 | #include <linux/err.h> | ||
30 | #include <linux/cpu.h> | ||
31 | #include <linux/smp.h> | ||
32 | #include <linux/seq_file.h> | ||
33 | #include <linux/irq.h> | ||
34 | #include <linux/percpu.h> | ||
35 | #include <linux/clockchips.h> | ||
36 | #include <linux/completion.h> | ||
37 | #include <linux/of.h> | ||
38 | |||
39 | #include <asm/atomic.h> | ||
40 | #include <asm/cacheflush.h> | ||
41 | #include <asm/cputype.h> | ||
42 | #include <asm/mmu_context.h> | ||
43 | #include <asm/pgtable.h> | ||
44 | #include <asm/pgalloc.h> | ||
45 | #include <asm/processor.h> | ||
46 | #include <asm/sections.h> | ||
47 | #include <asm/tlbflush.h> | ||
48 | #include <asm/ptrace.h> | ||
49 | #include <asm/mmu_context.h> | ||
50 | |||
51 | /* | ||
52 | * as from 2.5, kernels no longer have an init_tasks structure | ||
53 | * so we need some other way of telling a new secondary core | ||
54 | * where to place its SVC stack | ||
55 | */ | ||
56 | struct secondary_data secondary_data; | ||
57 | volatile unsigned long secondary_holding_pen_release = -1; | ||
58 | |||
59 | enum ipi_msg_type { | ||
60 | IPI_RESCHEDULE, | ||
61 | IPI_CALL_FUNC, | ||
62 | IPI_CALL_FUNC_SINGLE, | ||
63 | IPI_CPU_STOP, | ||
64 | }; | ||
65 | |||
66 | static DEFINE_RAW_SPINLOCK(boot_lock); | ||
67 | |||
68 | /* | ||
69 | * Write secondary_holding_pen_release in a way that is guaranteed to be | ||
70 | * visible to all observers, irrespective of whether they're taking part | ||
71 | * in coherency or not. This is necessary for the hotplug code to work | ||
72 | * reliably. | ||
73 | */ | ||
74 | static void __cpuinit write_pen_release(int val) | ||
75 | { | ||
76 | void *start = (void *)&secondary_holding_pen_release; | ||
77 | unsigned long size = sizeof(secondary_holding_pen_release); | ||
78 | |||
79 | secondary_holding_pen_release = val; | ||
80 | __flush_dcache_area(start, size); | ||
81 | } | ||
82 | |||
83 | /* | ||
84 | * Boot a secondary CPU, and assign it the specified idle task. | ||
85 | * This also gives us the initial stack to use for this CPU. | ||
86 | */ | ||
87 | static int __cpuinit boot_secondary(unsigned int cpu, struct task_struct *idle) | ||
88 | { | ||
89 | unsigned long timeout; | ||
90 | |||
91 | /* | ||
92 | * Set synchronisation state between this boot processor | ||
93 | * and the secondary one | ||
94 | */ | ||
95 | raw_spin_lock(&boot_lock); | ||
96 | |||
97 | /* | ||
98 | * Update the pen release flag. | ||
99 | */ | ||
100 | write_pen_release(cpu); | ||
101 | |||
102 | /* | ||
103 | * Send an event, causing the secondaries to read pen_release. | ||
104 | */ | ||
105 | sev(); | ||
106 | |||
107 | timeout = jiffies + (1 * HZ); | ||
108 | while (time_before(jiffies, timeout)) { | ||
109 | if (secondary_holding_pen_release == -1UL) | ||
110 | break; | ||
111 | udelay(10); | ||
112 | } | ||
113 | |||
114 | /* | ||
115 | * Now the secondary core is starting up let it run its | ||
116 | * calibrations, then wait for it to finish | ||
117 | */ | ||
118 | raw_spin_unlock(&boot_lock); | ||
119 | |||
120 | return secondary_holding_pen_release != -1 ? -ENOSYS : 0; | ||
121 | } | ||
122 | |||
123 | static DECLARE_COMPLETION(cpu_running); | ||
124 | |||
125 | int __cpuinit __cpu_up(unsigned int cpu, struct task_struct *idle) | ||
126 | { | ||
127 | int ret; | ||
128 | |||
129 | /* | ||
130 | * We need to tell the secondary core where to find its stack and the | ||
131 | * page tables. | ||
132 | */ | ||
133 | secondary_data.stack = task_stack_page(idle) + THREAD_START_SP; | ||
134 | __flush_dcache_area(&secondary_data, sizeof(secondary_data)); | ||
135 | |||
136 | /* | ||
137 | * Now bring the CPU into our world. | ||
138 | */ | ||
139 | ret = boot_secondary(cpu, idle); | ||
140 | if (ret == 0) { | ||
141 | /* | ||
142 | * CPU was successfully started, wait for it to come online or | ||
143 | * time out. | ||
144 | */ | ||
145 | wait_for_completion_timeout(&cpu_running, | ||
146 | msecs_to_jiffies(1000)); | ||
147 | |||
148 | if (!cpu_online(cpu)) { | ||
149 | pr_crit("CPU%u: failed to come online\n", cpu); | ||
150 | ret = -EIO; | ||
151 | } | ||
152 | } else { | ||
153 | pr_err("CPU%u: failed to boot: %d\n", cpu, ret); | ||
154 | } | ||
155 | |||
156 | secondary_data.stack = NULL; | ||
157 | |||
158 | return ret; | ||
159 | } | ||
160 | |||
161 | /* | ||
162 | * This is the secondary CPU boot entry. We're using this CPUs | ||
163 | * idle thread stack, but a set of temporary page tables. | ||
164 | */ | ||
165 | asmlinkage void __cpuinit secondary_start_kernel(void) | ||
166 | { | ||
167 | struct mm_struct *mm = &init_mm; | ||
168 | unsigned int cpu = smp_processor_id(); | ||
169 | |||
170 | printk("CPU%u: Booted secondary processor\n", cpu); | ||
171 | |||
172 | /* | ||
173 | * All kernel threads share the same mm context; grab a | ||
174 | * reference and switch to it. | ||
175 | */ | ||
176 | atomic_inc(&mm->mm_count); | ||
177 | current->active_mm = mm; | ||
178 | cpumask_set_cpu(cpu, mm_cpumask(mm)); | ||
179 | |||
180 | /* | ||
181 | * TTBR0 is only used for the identity mapping at this stage. Make it | ||
182 | * point to zero page to avoid speculatively fetching new entries. | ||
183 | */ | ||
184 | cpu_set_reserved_ttbr0(); | ||
185 | flush_tlb_all(); | ||
186 | |||
187 | preempt_disable(); | ||
188 | trace_hardirqs_off(); | ||
189 | |||
190 | /* | ||
191 | * Let the primary processor know we're out of the | ||
192 | * pen, then head off into the C entry point | ||
193 | */ | ||
194 | write_pen_release(-1); | ||
195 | |||
196 | /* | ||
197 | * Synchronise with the boot thread. | ||
198 | */ | ||
199 | raw_spin_lock(&boot_lock); | ||
200 | raw_spin_unlock(&boot_lock); | ||
201 | |||
202 | /* | ||
203 | * Enable local interrupts. | ||
204 | */ | ||
205 | notify_cpu_starting(cpu); | ||
206 | local_irq_enable(); | ||
207 | local_fiq_enable(); | ||
208 | |||
209 | /* | ||
210 | * OK, now it's safe to let the boot CPU continue. Wait for | ||
211 | * the CPU migration code to notice that the CPU is online | ||
212 | * before we continue. | ||
213 | */ | ||
214 | set_cpu_online(cpu, true); | ||
215 | while (!cpu_active(cpu)) | ||
216 | cpu_relax(); | ||
217 | |||
218 | /* | ||
219 | * OK, it's off to the idle thread for us | ||
220 | */ | ||
221 | cpu_idle(); | ||
222 | } | ||
223 | |||
224 | void __init smp_cpus_done(unsigned int max_cpus) | ||
225 | { | ||
226 | unsigned long bogosum = loops_per_jiffy * num_online_cpus(); | ||
227 | |||
228 | pr_info("SMP: Total of %d processors activated (%lu.%02lu BogoMIPS).\n", | ||
229 | num_online_cpus(), bogosum / (500000/HZ), | ||
230 | (bogosum / (5000/HZ)) % 100); | ||
231 | } | ||
232 | |||
233 | void __init smp_prepare_boot_cpu(void) | ||
234 | { | ||
235 | } | ||
236 | |||
237 | static void (*smp_cross_call)(const struct cpumask *, unsigned int); | ||
238 | static phys_addr_t cpu_release_addr[NR_CPUS]; | ||
239 | |||
240 | /* | ||
241 | * Enumerate the possible CPU set from the device tree. | ||
242 | */ | ||
243 | void __init smp_init_cpus(void) | ||
244 | { | ||
245 | const char *enable_method; | ||
246 | struct device_node *dn = NULL; | ||
247 | int cpu = 0; | ||
248 | |||
249 | while ((dn = of_find_node_by_type(dn, "cpu"))) { | ||
250 | if (cpu >= NR_CPUS) | ||
251 | goto next; | ||
252 | |||
253 | /* | ||
254 | * We currently support only the "spin-table" enable-method. | ||
255 | */ | ||
256 | enable_method = of_get_property(dn, "enable-method", NULL); | ||
257 | if (!enable_method || strcmp(enable_method, "spin-table")) { | ||
258 | pr_err("CPU %d: missing or invalid enable-method property: %s\n", | ||
259 | cpu, enable_method); | ||
260 | goto next; | ||
261 | } | ||
262 | |||
263 | /* | ||
264 | * Determine the address from which the CPU is polling. | ||
265 | */ | ||
266 | if (of_property_read_u64(dn, "cpu-release-addr", | ||
267 | &cpu_release_addr[cpu])) { | ||
268 | pr_err("CPU %d: missing or invalid cpu-release-addr property\n", | ||
269 | cpu); | ||
270 | goto next; | ||
271 | } | ||
272 | |||
273 | set_cpu_possible(cpu, true); | ||
274 | next: | ||
275 | cpu++; | ||
276 | } | ||
277 | |||
278 | /* sanity check */ | ||
279 | if (cpu > NR_CPUS) | ||
280 | pr_warning("no. of cores (%d) greater than configured maximum of %d - clipping\n", | ||
281 | cpu, NR_CPUS); | ||
282 | } | ||
283 | |||
284 | void __init smp_prepare_cpus(unsigned int max_cpus) | ||
285 | { | ||
286 | int cpu; | ||
287 | void **release_addr; | ||
288 | unsigned int ncores = num_possible_cpus(); | ||
289 | |||
290 | /* | ||
291 | * are we trying to boot more cores than exist? | ||
292 | */ | ||
293 | if (max_cpus > ncores) | ||
294 | max_cpus = ncores; | ||
295 | |||
296 | /* | ||
297 | * Initialise the present map (which describes the set of CPUs | ||
298 | * actually populated at the present time) and release the | ||
299 | * secondaries from the bootloader. | ||
300 | */ | ||
301 | for_each_possible_cpu(cpu) { | ||
302 | if (max_cpus == 0) | ||
303 | break; | ||
304 | |||
305 | if (!cpu_release_addr[cpu]) | ||
306 | continue; | ||
307 | |||
308 | release_addr = __va(cpu_release_addr[cpu]); | ||
309 | release_addr[0] = (void *)__pa(secondary_holding_pen); | ||
310 | __flush_dcache_area(release_addr, sizeof(release_addr[0])); | ||
311 | |||
312 | set_cpu_present(cpu, true); | ||
313 | max_cpus--; | ||
314 | } | ||
315 | |||
316 | /* | ||
317 | * Send an event to wake up the secondaries. | ||
318 | */ | ||
319 | sev(); | ||
320 | } | ||
321 | |||
322 | |||
323 | void __init set_smp_cross_call(void (*fn)(const struct cpumask *, unsigned int)) | ||
324 | { | ||
325 | smp_cross_call = fn; | ||
326 | } | ||
327 | |||
328 | void arch_send_call_function_ipi_mask(const struct cpumask *mask) | ||
329 | { | ||
330 | smp_cross_call(mask, IPI_CALL_FUNC); | ||
331 | } | ||
332 | |||
333 | void arch_send_call_function_single_ipi(int cpu) | ||
334 | { | ||
335 | smp_cross_call(cpumask_of(cpu), IPI_CALL_FUNC_SINGLE); | ||
336 | } | ||
337 | |||
338 | static const char *ipi_types[NR_IPI] = { | ||
339 | #define S(x,s) [x - IPI_RESCHEDULE] = s | ||
340 | S(IPI_RESCHEDULE, "Rescheduling interrupts"), | ||
341 | S(IPI_CALL_FUNC, "Function call interrupts"), | ||
342 | S(IPI_CALL_FUNC_SINGLE, "Single function call interrupts"), | ||
343 | S(IPI_CPU_STOP, "CPU stop interrupts"), | ||
344 | }; | ||
345 | |||
346 | void show_ipi_list(struct seq_file *p, int prec) | ||
347 | { | ||
348 | unsigned int cpu, i; | ||
349 | |||
350 | for (i = 0; i < NR_IPI; i++) { | ||
351 | seq_printf(p, "%*s%u:%s", prec - 1, "IPI", i + IPI_RESCHEDULE, | ||
352 | prec >= 4 ? " " : ""); | ||
353 | for_each_present_cpu(cpu) | ||
354 | seq_printf(p, "%10u ", | ||
355 | __get_irq_stat(cpu, ipi_irqs[i])); | ||
356 | seq_printf(p, " %s\n", ipi_types[i]); | ||
357 | } | ||
358 | } | ||
359 | |||
360 | u64 smp_irq_stat_cpu(unsigned int cpu) | ||
361 | { | ||
362 | u64 sum = 0; | ||
363 | int i; | ||
364 | |||
365 | for (i = 0; i < NR_IPI; i++) | ||
366 | sum += __get_irq_stat(cpu, ipi_irqs[i]); | ||
367 | |||
368 | return sum; | ||
369 | } | ||
370 | |||
371 | static DEFINE_RAW_SPINLOCK(stop_lock); | ||
372 | |||
373 | /* | ||
374 | * ipi_cpu_stop - handle IPI from smp_send_stop() | ||
375 | */ | ||
376 | static void ipi_cpu_stop(unsigned int cpu) | ||
377 | { | ||
378 | if (system_state == SYSTEM_BOOTING || | ||
379 | system_state == SYSTEM_RUNNING) { | ||
380 | raw_spin_lock(&stop_lock); | ||
381 | pr_crit("CPU%u: stopping\n", cpu); | ||
382 | dump_stack(); | ||
383 | raw_spin_unlock(&stop_lock); | ||
384 | } | ||
385 | |||
386 | set_cpu_online(cpu, false); | ||
387 | |||
388 | local_fiq_disable(); | ||
389 | local_irq_disable(); | ||
390 | |||
391 | while (1) | ||
392 | cpu_relax(); | ||
393 | } | ||
394 | |||
395 | /* | ||
396 | * Main handler for inter-processor interrupts | ||
397 | */ | ||
398 | void handle_IPI(int ipinr, struct pt_regs *regs) | ||
399 | { | ||
400 | unsigned int cpu = smp_processor_id(); | ||
401 | struct pt_regs *old_regs = set_irq_regs(regs); | ||
402 | |||
403 | if (ipinr >= IPI_RESCHEDULE && ipinr < IPI_RESCHEDULE + NR_IPI) | ||
404 | __inc_irq_stat(cpu, ipi_irqs[ipinr - IPI_RESCHEDULE]); | ||
405 | |||
406 | switch (ipinr) { | ||
407 | case IPI_RESCHEDULE: | ||
408 | scheduler_ipi(); | ||
409 | break; | ||
410 | |||
411 | case IPI_CALL_FUNC: | ||
412 | irq_enter(); | ||
413 | generic_smp_call_function_interrupt(); | ||
414 | irq_exit(); | ||
415 | break; | ||
416 | |||
417 | case IPI_CALL_FUNC_SINGLE: | ||
418 | irq_enter(); | ||
419 | generic_smp_call_function_single_interrupt(); | ||
420 | irq_exit(); | ||
421 | break; | ||
422 | |||
423 | case IPI_CPU_STOP: | ||
424 | irq_enter(); | ||
425 | ipi_cpu_stop(cpu); | ||
426 | irq_exit(); | ||
427 | break; | ||
428 | |||
429 | default: | ||
430 | pr_crit("CPU%u: Unknown IPI message 0x%x\n", cpu, ipinr); | ||
431 | break; | ||
432 | } | ||
433 | set_irq_regs(old_regs); | ||
434 | } | ||
435 | |||
436 | void smp_send_reschedule(int cpu) | ||
437 | { | ||
438 | smp_cross_call(cpumask_of(cpu), IPI_RESCHEDULE); | ||
439 | } | ||
440 | |||
441 | void smp_send_stop(void) | ||
442 | { | ||
443 | unsigned long timeout; | ||
444 | |||
445 | if (num_online_cpus() > 1) { | ||
446 | cpumask_t mask; | ||
447 | |||
448 | cpumask_copy(&mask, cpu_online_mask); | ||
449 | cpu_clear(smp_processor_id(), mask); | ||
450 | |||
451 | smp_cross_call(&mask, IPI_CPU_STOP); | ||
452 | } | ||
453 | |||
454 | /* Wait up to one second for other CPUs to stop */ | ||
455 | timeout = USEC_PER_SEC; | ||
456 | while (num_online_cpus() > 1 && timeout--) | ||
457 | udelay(1); | ||
458 | |||
459 | if (num_online_cpus() > 1) | ||
460 | pr_warning("SMP: failed to stop secondary CPUs\n"); | ||
461 | } | ||
462 | |||
463 | /* | ||
464 | * not supported here | ||
465 | */ | ||
466 | int setup_profiling_timer(unsigned int multiplier) | ||
467 | { | ||
468 | return -EINVAL; | ||
469 | } | ||