2 * This program is free software; you can redistribute it and/or
3 * modify it under the terms of the GNU General Public License
4 * as published by the Free Software Foundation; either version 2
5 * of the License, or (at your option) any later version.
7 * This program is distributed in the hope that it will be useful,
8 * but WITHOUT ANY WARRANTY; without even the implied warranty of
9 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
10 * GNU General Public License for more details.
12 * You should have received a copy of the GNU General Public License
13 * along with this program; if not, write to the Free Software
14 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
16 * Copyright (C) 2000, 2001 Kanoj Sarcar
17 * Copyright (C) 2000, 2001 Ralf Baechle
18 * Copyright (C) 2000, 2001 Silicon Graphics, Inc.
19 * Copyright (C) 2000, 2001, 2003 Broadcom Corporation
21 #include <linux/cache.h>
22 #include <linux/delay.h>
23 #include <linux/init.h>
24 #include <linux/interrupt.h>
25 #include <linux/smp.h>
26 #include <linux/spinlock.h>
27 #include <linux/threads.h>
28 #include <linux/export.h>
29 #include <linux/time.h>
30 #include <linux/timex.h>
31 #include <linux/sched/mm.h>
32 #include <linux/cpumask.h>
33 #include <linux/cpu.h>
34 #include <linux/err.h>
35 #include <linux/ftrace.h>
36 #include <linux/irqdomain.h>
38 #include <linux/of_irq.h>
40 #include <linux/atomic.h>
42 #include <asm/processor.h>
44 #include <asm/r4k-timer.h>
45 #include <asm/mips-cpc.h>
46 #include <asm/mmu_context.h>
48 #include <asm/setup.h>
51 int __cpu_number_map
[NR_CPUS
]; /* Map physical to logical */
52 EXPORT_SYMBOL(__cpu_number_map
);
54 int __cpu_logical_map
[NR_CPUS
]; /* Map logical to physical */
55 EXPORT_SYMBOL(__cpu_logical_map
);
57 /* Number of TCs (or siblings in Intel speak) per CPU core */
58 int smp_num_siblings
= 1;
59 EXPORT_SYMBOL(smp_num_siblings
);
61 /* representing the TCs (or siblings in Intel speak) of each logical CPU */
62 cpumask_t cpu_sibling_map
[NR_CPUS
] __read_mostly
;
63 EXPORT_SYMBOL(cpu_sibling_map
);
65 /* representing the core map of multi-core chips of each logical CPU */
66 cpumask_t cpu_core_map
[NR_CPUS
] __read_mostly
;
67 EXPORT_SYMBOL(cpu_core_map
);
69 static DECLARE_COMPLETION(cpu_running
);
72 * A logcal cpu mask containing only one VPE per core to
73 * reduce the number of IPIs on large MT systems.
75 cpumask_t cpu_foreign_map
[NR_CPUS
] __read_mostly
;
76 EXPORT_SYMBOL(cpu_foreign_map
);
78 /* representing cpus for which sibling maps can be computed */
79 static cpumask_t cpu_sibling_setup_map
;
81 /* representing cpus for which core maps can be computed */
82 static cpumask_t cpu_core_setup_map
;
84 cpumask_t cpu_coherent_mask
;
86 #ifdef CONFIG_GENERIC_IRQ_IPI
87 static struct irq_desc
*call_desc
;
88 static struct irq_desc
*sched_desc
;
91 static inline void set_cpu_sibling_map(int cpu
)
95 cpumask_set_cpu(cpu
, &cpu_sibling_setup_map
);
97 if (smp_num_siblings
> 1) {
98 for_each_cpu(i
, &cpu_sibling_setup_map
) {
99 if (cpu_data
[cpu
].package
== cpu_data
[i
].package
&&
100 cpu_data
[cpu
].core
== cpu_data
[i
].core
) {
101 cpumask_set_cpu(i
, &cpu_sibling_map
[cpu
]);
102 cpumask_set_cpu(cpu
, &cpu_sibling_map
[i
]);
106 cpumask_set_cpu(cpu
, &cpu_sibling_map
[cpu
]);
109 static inline void set_cpu_core_map(int cpu
)
113 cpumask_set_cpu(cpu
, &cpu_core_setup_map
);
115 for_each_cpu(i
, &cpu_core_setup_map
) {
116 if (cpu_data
[cpu
].package
== cpu_data
[i
].package
) {
117 cpumask_set_cpu(i
, &cpu_core_map
[cpu
]);
118 cpumask_set_cpu(cpu
, &cpu_core_map
[i
]);
124 * Calculate a new cpu_foreign_map mask whenever a
125 * new cpu appears or disappears.
127 void calculate_cpu_foreign_map(void)
129 int i
, k
, core_present
;
130 cpumask_t temp_foreign_map
;
132 /* Re-calculate the mask */
133 cpumask_clear(&temp_foreign_map
);
134 for_each_online_cpu(i
) {
136 for_each_cpu(k
, &temp_foreign_map
)
137 if (cpu_data
[i
].package
== cpu_data
[k
].package
&&
138 cpu_data
[i
].core
== cpu_data
[k
].core
)
141 cpumask_set_cpu(i
, &temp_foreign_map
);
144 for_each_online_cpu(i
)
145 cpumask_andnot(&cpu_foreign_map
[i
],
146 &temp_foreign_map
, &cpu_sibling_map
[i
]);
149 struct plat_smp_ops
*mp_ops
;
150 EXPORT_SYMBOL(mp_ops
);
152 void register_smp_ops(struct plat_smp_ops
*ops
)
155 printk(KERN_WARNING
"Overriding previously set SMP ops\n");
160 #ifdef CONFIG_GENERIC_IRQ_IPI
161 void mips_smp_send_ipi_single(int cpu
, unsigned int action
)
163 mips_smp_send_ipi_mask(cpumask_of(cpu
), action
);
166 void mips_smp_send_ipi_mask(const struct cpumask
*mask
, unsigned int action
)
172 local_irq_save(flags
);
175 case SMP_CALL_FUNCTION
:
176 __ipi_send_mask(call_desc
, mask
);
179 case SMP_RESCHEDULE_YOURSELF
:
180 __ipi_send_mask(sched_desc
, mask
);
187 if (mips_cpc_present()) {
188 for_each_cpu(cpu
, mask
) {
189 core
= cpu_data
[cpu
].core
;
191 if (core
== current_cpu_data
.core
)
194 while (!cpumask_test_cpu(cpu
, &cpu_coherent_mask
)) {
195 mips_cm_lock_other(core
, 0);
196 mips_cpc_lock_other(core
);
197 write_cpc_co_cmd(CPC_Cx_CMD_PWRUP
);
198 mips_cpc_unlock_other();
199 mips_cm_unlock_other();
204 local_irq_restore(flags
);
208 static irqreturn_t
ipi_resched_interrupt(int irq
, void *dev_id
)
215 static irqreturn_t
ipi_call_interrupt(int irq
, void *dev_id
)
217 generic_smp_call_function_interrupt();
222 static struct irqaction irq_resched
= {
223 .handler
= ipi_resched_interrupt
,
224 .flags
= IRQF_PERCPU
,
225 .name
= "IPI resched"
228 static struct irqaction irq_call
= {
229 .handler
= ipi_call_interrupt
,
230 .flags
= IRQF_PERCPU
,
234 static void smp_ipi_init_one(unsigned int virq
,
235 struct irqaction
*action
)
239 irq_set_handler(virq
, handle_percpu_irq
);
240 ret
= setup_irq(virq
, action
);
244 static unsigned int call_virq
, sched_virq
;
246 int mips_smp_ipi_allocate(const struct cpumask
*mask
)
249 struct irq_domain
*ipidomain
;
250 struct device_node
*node
;
252 node
= of_irq_find_parent(of_root
);
253 ipidomain
= irq_find_matching_host(node
, DOMAIN_BUS_IPI
);
256 * Some platforms have half DT setup. So if we found irq node but
257 * didn't find an ipidomain, try to search for one that is not in the
260 if (node
&& !ipidomain
)
261 ipidomain
= irq_find_matching_host(NULL
, DOMAIN_BUS_IPI
);
264 * There are systems which use IPI IRQ domains, but only have one
265 * registered when some runtime condition is met. For example a Malta
266 * kernel may include support for GIC & CPU interrupt controller IPI
267 * IRQ domains, but if run on a system with no GIC & no MT ASE then
268 * neither will be supported or registered.
270 * We only have a problem if we're actually using multiple CPUs so fail
271 * loudly if that is the case. Otherwise simply return, skipping IPI
272 * setup, if we're running with only a single CPU.
275 BUG_ON(num_present_cpus() > 1);
279 virq
= irq_reserve_ipi(ipidomain
, mask
);
284 virq
= irq_reserve_ipi(ipidomain
, mask
);
289 if (irq_domain_is_ipi_per_cpu(ipidomain
)) {
292 for_each_cpu(cpu
, mask
) {
293 smp_ipi_init_one(call_virq
+ cpu
, &irq_call
);
294 smp_ipi_init_one(sched_virq
+ cpu
, &irq_resched
);
297 smp_ipi_init_one(call_virq
, &irq_call
);
298 smp_ipi_init_one(sched_virq
, &irq_resched
);
304 int mips_smp_ipi_free(const struct cpumask
*mask
)
306 struct irq_domain
*ipidomain
;
307 struct device_node
*node
;
309 node
= of_irq_find_parent(of_root
);
310 ipidomain
= irq_find_matching_host(node
, DOMAIN_BUS_IPI
);
313 * Some platforms have half DT setup. So if we found irq node but
314 * didn't find an ipidomain, try to search for one that is not in the
317 if (node
&& !ipidomain
)
318 ipidomain
= irq_find_matching_host(NULL
, DOMAIN_BUS_IPI
);
322 if (irq_domain_is_ipi_per_cpu(ipidomain
)) {
325 for_each_cpu(cpu
, mask
) {
326 remove_irq(call_virq
+ cpu
, &irq_call
);
327 remove_irq(sched_virq
+ cpu
, &irq_resched
);
330 irq_destroy_ipi(call_virq
, mask
);
331 irq_destroy_ipi(sched_virq
, mask
);
336 static int __init
mips_smp_ipi_init(void)
338 if (num_possible_cpus() == 1)
341 mips_smp_ipi_allocate(cpu_possible_mask
);
343 call_desc
= irq_to_desc(call_virq
);
344 sched_desc
= irq_to_desc(sched_virq
);
348 early_initcall(mips_smp_ipi_init
);
352 * First C code run on the secondary CPUs after being started up by
355 asmlinkage
void start_secondary(void)
360 per_cpu_trap_init(false);
361 mips_clockevent_init();
362 mp_ops
->init_secondary();
367 * XXX parity protection should be folded in here when it's converted
368 * to an option instead of something based on .cputype
373 cpu
= smp_processor_id();
374 cpu_data
[cpu
].udelay_val
= loops_per_jiffy
;
376 cpumask_set_cpu(cpu
, &cpu_coherent_mask
);
377 notify_cpu_starting(cpu
);
379 set_cpu_online(cpu
, true);
381 set_cpu_sibling_map(cpu
);
382 set_cpu_core_map(cpu
);
384 calculate_cpu_foreign_map();
386 complete(&cpu_running
);
387 synchronise_count_slave(cpu
);
390 * irq will be enabled in ->smp_finish(), enabling it too early
393 WARN_ON_ONCE(!irqs_disabled());
394 mp_ops
->smp_finish();
396 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE
);
399 static void stop_this_cpu(void *dummy
)
405 set_cpu_online(smp_processor_id(), false);
406 calculate_cpu_foreign_map();
411 void smp_send_stop(void)
413 smp_call_function(stop_this_cpu
, NULL
, 0);
416 void __init
smp_cpus_done(unsigned int max_cpus
)
420 /* called from main before smp_init() */
421 void __init
smp_prepare_cpus(unsigned int max_cpus
)
423 init_new_context(current
, &init_mm
);
424 current_thread_info()->cpu
= 0;
425 mp_ops
->prepare_cpus(max_cpus
);
426 set_cpu_sibling_map(0);
428 calculate_cpu_foreign_map();
429 #ifndef CONFIG_HOTPLUG_CPU
430 init_cpu_present(cpu_possible_mask
);
432 cpumask_copy(&cpu_coherent_mask
, cpu_possible_mask
);
435 /* preload SMP state for boot cpu */
436 void smp_prepare_boot_cpu(void)
438 set_cpu_possible(0, true);
439 set_cpu_online(0, true);
442 int __cpu_up(unsigned int cpu
, struct task_struct
*tidle
)
444 mp_ops
->boot_secondary(cpu
, tidle
);
447 * We must check for timeout here, as the CPU will not be marked
448 * online until the counters are synchronised.
450 if (!wait_for_completion_timeout(&cpu_running
,
451 msecs_to_jiffies(1000))) {
452 pr_crit("CPU%u: failed to start\n", cpu
);
456 synchronise_count_master(cpu
);
460 /* Not really SMP stuff ... */
461 int setup_profiling_timer(unsigned int multiplier
)
466 static void flush_tlb_all_ipi(void *info
)
468 local_flush_tlb_all();
471 void flush_tlb_all(void)
473 on_each_cpu(flush_tlb_all_ipi
, NULL
, 1);
476 static void flush_tlb_mm_ipi(void *mm
)
478 local_flush_tlb_mm((struct mm_struct
*)mm
);
482 * Special Variant of smp_call_function for use by TLB functions:
485 * o collapses to normal function call on UP kernels
486 * o collapses to normal function call on systems with a single shared
489 static inline void smp_on_other_tlbs(void (*func
) (void *info
), void *info
)
491 smp_call_function(func
, info
, 1);
494 static inline void smp_on_each_tlb(void (*func
) (void *info
), void *info
)
498 smp_on_other_tlbs(func
, info
);
505 * The following tlb flush calls are invoked when old translations are
506 * being torn down, or pte attributes are changing. For single threaded
507 * address spaces, a new context is obtained on the current cpu, and tlb
508 * context on other cpus are invalidated to force a new context allocation
509 * at switch_mm time, should the mm ever be used on other cpus. For
510 * multithreaded address spaces, intercpu interrupts have to be sent.
511 * Another case where intercpu interrupts are required is when the target
512 * mm might be active on another cpu (eg debuggers doing the flushes on
513 * behalf of debugees, kswapd stealing pages from another process etc).
517 void flush_tlb_mm(struct mm_struct
*mm
)
521 if ((atomic_read(&mm
->mm_users
) != 1) || (current
->mm
!= mm
)) {
522 smp_on_other_tlbs(flush_tlb_mm_ipi
, mm
);
526 for_each_online_cpu(cpu
) {
527 if (cpu
!= smp_processor_id() && cpu_context(cpu
, mm
))
528 cpu_context(cpu
, mm
) = 0;
531 local_flush_tlb_mm(mm
);
536 struct flush_tlb_data
{
537 struct vm_area_struct
*vma
;
542 static void flush_tlb_range_ipi(void *info
)
544 struct flush_tlb_data
*fd
= info
;
546 local_flush_tlb_range(fd
->vma
, fd
->addr1
, fd
->addr2
);
549 void flush_tlb_range(struct vm_area_struct
*vma
, unsigned long start
, unsigned long end
)
551 struct mm_struct
*mm
= vma
->vm_mm
;
554 if ((atomic_read(&mm
->mm_users
) != 1) || (current
->mm
!= mm
)) {
555 struct flush_tlb_data fd
= {
561 smp_on_other_tlbs(flush_tlb_range_ipi
, &fd
);
564 int exec
= vma
->vm_flags
& VM_EXEC
;
566 for_each_online_cpu(cpu
) {
568 * flush_cache_range() will only fully flush icache if
569 * the VMA is executable, otherwise we must invalidate
570 * ASID without it appearing to has_valid_asid() as if
571 * mm has been completely unused by that CPU.
573 if (cpu
!= smp_processor_id() && cpu_context(cpu
, mm
))
574 cpu_context(cpu
, mm
) = !exec
;
577 local_flush_tlb_range(vma
, start
, end
);
581 static void flush_tlb_kernel_range_ipi(void *info
)
583 struct flush_tlb_data
*fd
= info
;
585 local_flush_tlb_kernel_range(fd
->addr1
, fd
->addr2
);
588 void flush_tlb_kernel_range(unsigned long start
, unsigned long end
)
590 struct flush_tlb_data fd
= {
595 on_each_cpu(flush_tlb_kernel_range_ipi
, &fd
, 1);
598 static void flush_tlb_page_ipi(void *info
)
600 struct flush_tlb_data
*fd
= info
;
602 local_flush_tlb_page(fd
->vma
, fd
->addr1
);
605 void flush_tlb_page(struct vm_area_struct
*vma
, unsigned long page
)
608 if ((atomic_read(&vma
->vm_mm
->mm_users
) != 1) || (current
->mm
!= vma
->vm_mm
)) {
609 struct flush_tlb_data fd
= {
614 smp_on_other_tlbs(flush_tlb_page_ipi
, &fd
);
618 for_each_online_cpu(cpu
) {
620 * flush_cache_page() only does partial flushes, so
621 * invalidate ASID without it appearing to
622 * has_valid_asid() as if mm has been completely unused
625 if (cpu
!= smp_processor_id() && cpu_context(cpu
, vma
->vm_mm
))
626 cpu_context(cpu
, vma
->vm_mm
) = 1;
629 local_flush_tlb_page(vma
, page
);
633 static void flush_tlb_one_ipi(void *info
)
635 unsigned long vaddr
= (unsigned long) info
;
637 local_flush_tlb_one(vaddr
);
640 void flush_tlb_one(unsigned long vaddr
)
642 smp_on_each_tlb(flush_tlb_one_ipi
, (void *) vaddr
);
645 EXPORT_SYMBOL(flush_tlb_page
);
646 EXPORT_SYMBOL(flush_tlb_one
);
648 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
650 static DEFINE_PER_CPU(atomic_t
, tick_broadcast_count
);
651 static DEFINE_PER_CPU(struct call_single_data
, tick_broadcast_csd
);
653 void tick_broadcast(const struct cpumask
*mask
)
656 struct call_single_data
*csd
;
659 for_each_cpu(cpu
, mask
) {
660 count
= &per_cpu(tick_broadcast_count
, cpu
);
661 csd
= &per_cpu(tick_broadcast_csd
, cpu
);
663 if (atomic_inc_return(count
) == 1)
664 smp_call_function_single_async(cpu
, csd
);
668 static void tick_broadcast_callee(void *info
)
670 int cpu
= smp_processor_id();
671 tick_receive_broadcast();
672 atomic_set(&per_cpu(tick_broadcast_count
, cpu
), 0);
675 static int __init
tick_broadcast_init(void)
677 struct call_single_data
*csd
;
680 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
681 csd
= &per_cpu(tick_broadcast_csd
, cpu
);
682 csd
->func
= tick_broadcast_callee
;
687 early_initcall(tick_broadcast_init
);
689 #endif /* CONFIG_GENERIC_CLOCKEVENTS_BROADCAST */