aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
-rw-r--r--arch/arm/kernel/smp.c70
-rw-r--r--arch/arm64/kernel/smp.c65
-rw-r--r--include/linux/ftrace_event.h34
-rw-r--r--include/linux/tracepoint.h44
-rw-r--r--include/trace/events/ipi.h89
5 files changed, 214 insertions, 88 deletions
diff --git a/arch/arm/kernel/smp.c b/arch/arm/kernel/smp.c
index 7c4fada440f0..9388a3d479e1 100644
--- a/arch/arm/kernel/smp.c
+++ b/arch/arm/kernel/smp.c
@@ -47,6 +47,9 @@
47#include <asm/mach/arch.h> 47#include <asm/mach/arch.h>
48#include <asm/mpu.h> 48#include <asm/mpu.h>
49 49
50#define CREATE_TRACE_POINTS
51#include <trace/events/ipi.h>
52
50/* 53/*
51 * as from 2.5, kernels no longer have an init_tasks structure 54 * as from 2.5, kernels no longer have an init_tasks structure
52 * so we need some other way of telling a new secondary core 55 * so we need some other way of telling a new secondary core
@@ -430,38 +433,15 @@ void __init smp_prepare_cpus(unsigned int max_cpus)
430 } 433 }
431} 434}
432 435
433static void (*smp_cross_call)(const struct cpumask *, unsigned int); 436static void (*__smp_cross_call)(const struct cpumask *, unsigned int);
434 437
435void __init set_smp_cross_call(void (*fn)(const struct cpumask *, unsigned int)) 438void __init set_smp_cross_call(void (*fn)(const struct cpumask *, unsigned int))
436{ 439{
437 if (!smp_cross_call) 440 if (!__smp_cross_call)
438 smp_cross_call = fn; 441 __smp_cross_call = fn;
439}
440
441void arch_send_call_function_ipi_mask(const struct cpumask *mask)
442{
443 smp_cross_call(mask, IPI_CALL_FUNC);
444}
445
446void arch_send_wakeup_ipi_mask(const struct cpumask *mask)
447{
448 smp_cross_call(mask, IPI_WAKEUP);
449}
450
451void arch_send_call_function_single_ipi(int cpu)
452{
453 smp_cross_call(cpumask_of(cpu), IPI_CALL_FUNC_SINGLE);
454} 442}
455 443
456#ifdef CONFIG_IRQ_WORK 444static const char *ipi_types[NR_IPI] __tracepoint_string = {
457void arch_irq_work_raise(void)
458{
459 if (is_smp())
460 smp_cross_call(cpumask_of(smp_processor_id()), IPI_IRQ_WORK);
461}
462#endif
463
464static const char *ipi_types[NR_IPI] = {
465#define S(x,s) [x] = s 445#define S(x,s) [x] = s
466 S(IPI_WAKEUP, "CPU wakeup interrupts"), 446 S(IPI_WAKEUP, "CPU wakeup interrupts"),
467 S(IPI_TIMER, "Timer broadcast interrupts"), 447 S(IPI_TIMER, "Timer broadcast interrupts"),
@@ -473,6 +453,12 @@ static const char *ipi_types[NR_IPI] = {
473 S(IPI_COMPLETION, "completion interrupts"), 453 S(IPI_COMPLETION, "completion interrupts"),
474}; 454};
475 455
456static void smp_cross_call(const struct cpumask *target, unsigned int ipinr)
457{
458 trace_ipi_raise(target, ipi_types[ipinr]);
459 __smp_cross_call(target, ipinr);
460}
461
476void show_ipi_list(struct seq_file *p, int prec) 462void show_ipi_list(struct seq_file *p, int prec)
477{ 463{
478 unsigned int cpu, i; 464 unsigned int cpu, i;
@@ -499,6 +485,29 @@ u64 smp_irq_stat_cpu(unsigned int cpu)
499 return sum; 485 return sum;
500} 486}
501 487
488void arch_send_call_function_ipi_mask(const struct cpumask *mask)
489{
490 smp_cross_call(mask, IPI_CALL_FUNC);
491}
492
493void arch_send_wakeup_ipi_mask(const struct cpumask *mask)
494{
495 smp_cross_call(mask, IPI_WAKEUP);
496}
497
498void arch_send_call_function_single_ipi(int cpu)
499{
500 smp_cross_call(cpumask_of(cpu), IPI_CALL_FUNC_SINGLE);
501}
502
503#ifdef CONFIG_IRQ_WORK
504void arch_irq_work_raise(void)
505{
506 if (is_smp())
507 smp_cross_call(cpumask_of(smp_processor_id()), IPI_IRQ_WORK);
508}
509#endif
510
502#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST 511#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
503void tick_broadcast(const struct cpumask *mask) 512void tick_broadcast(const struct cpumask *mask)
504{ 513{
@@ -556,8 +565,10 @@ void handle_IPI(int ipinr, struct pt_regs *regs)
556 unsigned int cpu = smp_processor_id(); 565 unsigned int cpu = smp_processor_id();
557 struct pt_regs *old_regs = set_irq_regs(regs); 566 struct pt_regs *old_regs = set_irq_regs(regs);
558 567
559 if (ipinr < NR_IPI) 568 if ((unsigned)ipinr < NR_IPI) {
569 trace_ipi_entry(ipi_types[ipinr]);
560 __inc_irq_stat(cpu, ipi_irqs[ipinr]); 570 __inc_irq_stat(cpu, ipi_irqs[ipinr]);
571 }
561 572
562 switch (ipinr) { 573 switch (ipinr) {
563 case IPI_WAKEUP: 574 case IPI_WAKEUP:
@@ -612,6 +623,9 @@ void handle_IPI(int ipinr, struct pt_regs *regs)
612 cpu, ipinr); 623 cpu, ipinr);
613 break; 624 break;
614 } 625 }
626
627 if ((unsigned)ipinr < NR_IPI)
628 trace_ipi_exit(ipi_types[ipinr]);
615 set_irq_regs(old_regs); 629 set_irq_regs(old_regs);
616} 630}
617 631
diff --git a/arch/arm64/kernel/smp.c b/arch/arm64/kernel/smp.c
index 3e2f5ebbf63e..474339718105 100644
--- a/arch/arm64/kernel/smp.c
+++ b/arch/arm64/kernel/smp.c
@@ -51,6 +51,9 @@
51#include <asm/tlbflush.h> 51#include <asm/tlbflush.h>
52#include <asm/ptrace.h> 52#include <asm/ptrace.h>
53 53
54#define CREATE_TRACE_POINTS
55#include <trace/events/ipi.h>
56
54/* 57/*
55 * as from 2.5, kernels no longer have an init_tasks structure 58 * as from 2.5, kernels no longer have an init_tasks structure
56 * so we need some other way of telling a new secondary core 59 * so we need some other way of telling a new secondary core
@@ -313,8 +316,6 @@ void __init smp_prepare_boot_cpu(void)
313 set_my_cpu_offset(per_cpu_offset(smp_processor_id())); 316 set_my_cpu_offset(per_cpu_offset(smp_processor_id()));
314} 317}
315 318
316static void (*smp_cross_call)(const struct cpumask *, unsigned int);
317
318/* 319/*
319 * Enumerate the possible CPU set from the device tree and build the 320 * Enumerate the possible CPU set from the device tree and build the
320 * cpu logical map array containing MPIDR values related to logical 321 * cpu logical map array containing MPIDR values related to logical
@@ -469,32 +470,15 @@ void __init smp_prepare_cpus(unsigned int max_cpus)
469 } 470 }
470} 471}
471 472
473static void (*__smp_cross_call)(const struct cpumask *, unsigned int);
472 474
473void __init set_smp_cross_call(void (*fn)(const struct cpumask *, unsigned int)) 475void __init set_smp_cross_call(void (*fn)(const struct cpumask *, unsigned int))
474{ 476{
475 smp_cross_call = fn; 477 __smp_cross_call = fn;
476} 478}
477 479
478void arch_send_call_function_ipi_mask(const struct cpumask *mask) 480static const char *ipi_types[NR_IPI] __tracepoint_string = {
479{ 481#define S(x,s) [x] = s
480 smp_cross_call(mask, IPI_CALL_FUNC);
481}
482
483void arch_send_call_function_single_ipi(int cpu)
484{
485 smp_cross_call(cpumask_of(cpu), IPI_CALL_FUNC_SINGLE);
486}
487
488#ifdef CONFIG_IRQ_WORK
489void arch_irq_work_raise(void)
490{
491 if (smp_cross_call)
492 smp_cross_call(cpumask_of(smp_processor_id()), IPI_IRQ_WORK);
493}
494#endif
495
496static const char *ipi_types[NR_IPI] = {
497#define S(x,s) [x - IPI_RESCHEDULE] = s
498 S(IPI_RESCHEDULE, "Rescheduling interrupts"), 482 S(IPI_RESCHEDULE, "Rescheduling interrupts"),
499 S(IPI_CALL_FUNC, "Function call interrupts"), 483 S(IPI_CALL_FUNC, "Function call interrupts"),
500 S(IPI_CALL_FUNC_SINGLE, "Single function call interrupts"), 484 S(IPI_CALL_FUNC_SINGLE, "Single function call interrupts"),
@@ -503,12 +487,18 @@ static const char *ipi_types[NR_IPI] = {
503 S(IPI_IRQ_WORK, "IRQ work interrupts"), 487 S(IPI_IRQ_WORK, "IRQ work interrupts"),
504}; 488};
505 489
490static void smp_cross_call(const struct cpumask *target, unsigned int ipinr)
491{
492 trace_ipi_raise(target, ipi_types[ipinr]);
493 __smp_cross_call(target, ipinr);
494}
495
506void show_ipi_list(struct seq_file *p, int prec) 496void show_ipi_list(struct seq_file *p, int prec)
507{ 497{
508 unsigned int cpu, i; 498 unsigned int cpu, i;
509 499
510 for (i = 0; i < NR_IPI; i++) { 500 for (i = 0; i < NR_IPI; i++) {
511 seq_printf(p, "%*s%u:%s", prec - 1, "IPI", i + IPI_RESCHEDULE, 501 seq_printf(p, "%*s%u:%s", prec - 1, "IPI", i,
512 prec >= 4 ? " " : ""); 502 prec >= 4 ? " " : "");
513 for_each_online_cpu(cpu) 503 for_each_online_cpu(cpu)
514 seq_printf(p, "%10u ", 504 seq_printf(p, "%10u ",
@@ -528,6 +518,24 @@ u64 smp_irq_stat_cpu(unsigned int cpu)
528 return sum; 518 return sum;
529} 519}
530 520
521void arch_send_call_function_ipi_mask(const struct cpumask *mask)
522{
523 smp_cross_call(mask, IPI_CALL_FUNC);
524}
525
526void arch_send_call_function_single_ipi(int cpu)
527{
528 smp_cross_call(cpumask_of(cpu), IPI_CALL_FUNC_SINGLE);
529}
530
531#ifdef CONFIG_IRQ_WORK
532void arch_irq_work_raise(void)
533{
534 if (__smp_cross_call)
535 smp_cross_call(cpumask_of(smp_processor_id()), IPI_IRQ_WORK);
536}
537#endif
538
531static DEFINE_RAW_SPINLOCK(stop_lock); 539static DEFINE_RAW_SPINLOCK(stop_lock);
532 540
533/* 541/*
@@ -559,8 +567,10 @@ void handle_IPI(int ipinr, struct pt_regs *regs)
559 unsigned int cpu = smp_processor_id(); 567 unsigned int cpu = smp_processor_id();
560 struct pt_regs *old_regs = set_irq_regs(regs); 568 struct pt_regs *old_regs = set_irq_regs(regs);
561 569
562 if (ipinr >= IPI_RESCHEDULE && ipinr < IPI_RESCHEDULE + NR_IPI) 570 if ((unsigned)ipinr < NR_IPI) {
563 __inc_irq_stat(cpu, ipi_irqs[ipinr - IPI_RESCHEDULE]); 571 trace_ipi_entry(ipi_types[ipinr]);
572 __inc_irq_stat(cpu, ipi_irqs[ipinr]);
573 }
564 574
565 switch (ipinr) { 575 switch (ipinr) {
566 case IPI_RESCHEDULE: 576 case IPI_RESCHEDULE:
@@ -605,6 +615,9 @@ void handle_IPI(int ipinr, struct pt_regs *regs)
605 pr_crit("CPU%u: Unknown IPI message 0x%x\n", cpu, ipinr); 615 pr_crit("CPU%u: Unknown IPI message 0x%x\n", cpu, ipinr);
606 break; 616 break;
607 } 617 }
618
619 if ((unsigned)ipinr < NR_IPI)
620 trace_ipi_exit(ipi_types[ipinr]);
608 set_irq_regs(old_regs); 621 set_irq_regs(old_regs);
609} 622}
610 623
diff --git a/include/linux/ftrace_event.h b/include/linux/ftrace_event.h
index 06c6faa9e5cc..28672e87e910 100644
--- a/include/linux/ftrace_event.h
+++ b/include/linux/ftrace_event.h
@@ -571,40 +571,6 @@ do { \
571 __trace_printk(ip, fmt, ##args); \ 571 __trace_printk(ip, fmt, ##args); \
572} while (0) 572} while (0)
573 573
574/**
575 * tracepoint_string - register constant persistent string to trace system
576 * @str - a constant persistent string that will be referenced in tracepoints
577 *
578 * If constant strings are being used in tracepoints, it is faster and
579 * more efficient to just save the pointer to the string and reference
580 * that with a printf "%s" instead of saving the string in the ring buffer
581 * and wasting space and time.
582 *
583 * The problem with the above approach is that userspace tools that read
584 * the binary output of the trace buffers do not have access to the string.
585 * Instead they just show the address of the string which is not very
586 * useful to users.
587 *
588 * With tracepoint_string(), the string will be registered to the tracing
589 * system and exported to userspace via the debugfs/tracing/printk_formats
590 * file that maps the string address to the string text. This way userspace
591 * tools that read the binary buffers have a way to map the pointers to
592 * the ASCII strings they represent.
593 *
594 * The @str used must be a constant string and persistent as it would not
595 * make sense to show a string that no longer exists. But it is still fine
596 * to be used with modules, because when modules are unloaded, if they
597 * had tracepoints, the ring buffers are cleared too. As long as the string
598 * does not change during the life of the module, it is fine to use
599 * tracepoint_string() within a module.
600 */
601#define tracepoint_string(str) \
602 ({ \
603 static const char *___tp_str __tracepoint_string = str; \
604 ___tp_str; \
605 })
606#define __tracepoint_string __attribute__((section("__tracepoint_str")))
607
608#ifdef CONFIG_PERF_EVENTS 574#ifdef CONFIG_PERF_EVENTS
609struct perf_event; 575struct perf_event;
610 576
diff --git a/include/linux/tracepoint.h b/include/linux/tracepoint.h
index 2e2a5f7717e5..b1293f15f592 100644
--- a/include/linux/tracepoint.h
+++ b/include/linux/tracepoint.h
@@ -249,6 +249,50 @@ extern void syscall_unregfunc(void);
249 249
250#endif /* CONFIG_TRACEPOINTS */ 250#endif /* CONFIG_TRACEPOINTS */
251 251
252#ifdef CONFIG_TRACING
253/**
254 * tracepoint_string - register constant persistent string to trace system
255 * @str - a constant persistent string that will be referenced in tracepoints
256 *
257 * If constant strings are being used in tracepoints, it is faster and
258 * more efficient to just save the pointer to the string and reference
259 * that with a printf "%s" instead of saving the string in the ring buffer
260 * and wasting space and time.
261 *
262 * The problem with the above approach is that userspace tools that read
263 * the binary output of the trace buffers do not have access to the string.
264 * Instead they just show the address of the string which is not very
265 * useful to users.
266 *
267 * With tracepoint_string(), the string will be registered to the tracing
268 * system and exported to userspace via the debugfs/tracing/printk_formats
269 * file that maps the string address to the string text. This way userspace
270 * tools that read the binary buffers have a way to map the pointers to
271 * the ASCII strings they represent.
272 *
273 * The @str used must be a constant string and persistent as it would not
274 * make sense to show a string that no longer exists. But it is still fine
275 * to be used with modules, because when modules are unloaded, if they
276 * had tracepoints, the ring buffers are cleared too. As long as the string
277 * does not change during the life of the module, it is fine to use
278 * tracepoint_string() within a module.
279 */
280#define tracepoint_string(str) \
281 ({ \
282 static const char *___tp_str __tracepoint_string = str; \
283 ___tp_str; \
284 })
285#define __tracepoint_string __attribute__((section("__tracepoint_str")))
286#else
287/*
288 * tracepoint_string() is used to save the string address for userspace
289 * tracing tools. When tracing isn't configured, there's no need to save
290 * anything.
291 */
292# define tracepoint_string(str) str
293# define __tracepoint_string
294#endif
295
252/* 296/*
253 * The need for the DECLARE_TRACE_NOARGS() is to handle the prototype 297 * The need for the DECLARE_TRACE_NOARGS() is to handle the prototype
254 * (void). "void" is a special value in a function prototype and can 298 * (void). "void" is a special value in a function prototype and can
diff --git a/include/trace/events/ipi.h b/include/trace/events/ipi.h
new file mode 100644
index 000000000000..834a7362a610
--- /dev/null
+++ b/include/trace/events/ipi.h
@@ -0,0 +1,89 @@
1#undef TRACE_SYSTEM
2#define TRACE_SYSTEM ipi
3
4#if !defined(_TRACE_IPI_H) || defined(TRACE_HEADER_MULTI_READ)
5#define _TRACE_IPI_H
6
7#include <linux/tracepoint.h>
8
9/**
10 * ipi_raise - called when a smp cross call is made
11 *
12 * @mask: mask of recipient CPUs for the IPI
13 * @reason: string identifying the IPI purpose
14 *
15 * It is necessary for @reason to be a static string declared with
16 * __tracepoint_string.
17 */
18TRACE_EVENT(ipi_raise,
19
20 TP_PROTO(const struct cpumask *mask, const char *reason),
21
22 TP_ARGS(mask, reason),
23
24 TP_STRUCT__entry(
25 __bitmask(target_cpus, nr_cpumask_bits)
26 __field(const char *, reason)
27 ),
28
29 TP_fast_assign(
30 __assign_bitmask(target_cpus, cpumask_bits(mask), nr_cpumask_bits);
31 __entry->reason = reason;
32 ),
33
34 TP_printk("target_mask=%s (%s)", __get_bitmask(target_cpus), __entry->reason)
35);
36
37DECLARE_EVENT_CLASS(ipi_handler,
38
39 TP_PROTO(const char *reason),
40
41 TP_ARGS(reason),
42
43 TP_STRUCT__entry(
44 __field(const char *, reason)
45 ),
46
47 TP_fast_assign(
48 __entry->reason = reason;
49 ),
50
51 TP_printk("(%s)", __entry->reason)
52);
53
54/**
55 * ipi_entry - called immediately before the IPI handler
56 *
57 * @reason: string identifying the IPI purpose
58 *
59 * It is necessary for @reason to be a static string declared with
60 * __tracepoint_string, ideally the same as used with trace_ipi_raise
61 * for that IPI.
62 */
63DEFINE_EVENT(ipi_handler, ipi_entry,
64
65 TP_PROTO(const char *reason),
66
67 TP_ARGS(reason)
68);
69
70/**
71 * ipi_exit - called immediately after the IPI handler returns
72 *
73 * @reason: string identifying the IPI purpose
74 *
75 * It is necessary for @reason to be a static string declared with
76 * __tracepoint_string, ideally the same as used with trace_ipi_raise for
77 * that IPI.
78 */
79DEFINE_EVENT(ipi_handler, ipi_exit,
80
81 TP_PROTO(const char *reason),
82
83 TP_ARGS(reason)
84);
85
86#endif /* _TRACE_IPI_H */
87
88/* This part must be outside protection */
89#include <trace/define_trace.h>