Merge branch 'perf-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip

* 'perf-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: perf session: Make events_stats u64 to avoid overflow on 32-bit arches hw-breakpoints: Fix hardware breakpoints -> perf events dependency perf events: Dont report side-band events on each cpu for per-task-per-cpu events perf events, x86/stacktrace: Fix performance/softlockup by providing a special frame pointer-only stack walker perf events, x86/stacktrace: Make stack walking optional perf events: Remove unused perf_counter.h header file perf probe: Check new event name kprobe-tracer: Check new event/group name perf probe: Check whether debugfs path is correct perf probe: Fix libdwarf include path for Debian

Linus Torvalds committed 16 years ago

Commit eca9dfc2 parents 3981e15 b5b60fd20 files · +175 −523Browse files

Unified · Split

20 files

arch/Kconfig +1 −3
arch/x86/Kconfig +2 −0
arch/x86/include/asm/stacktrace.h +24 −0
arch/x86/kernel/cpu/perf_event.c +1 −0
arch/x86/kernel/dumpstack.c +29 −4
arch/x86/kernel/dumpstack.h +0 −6
arch/x86/kernel/dumpstack_32.c +1 −1
arch/x86/kernel/dumpstack_64.c +2 −2
arch/x86/kernel/stacktrace.c +10 −8
arch/x86/oprofile/backtrace.c +5 −4
include/linux/perf_counter.h +0 −444
kernel/perf_event.c +18 −14
kernel/trace/trace_kprobe.c +25 −6
kernel/trace/trace_sysprof.c +1 −0
tools/perf/Makefile +2 −1
tools/perf/builtin-probe.c +4 −0
tools/perf/builtin-report.c +1 −1
tools/perf/util/event.h +2 −2
tools/perf/util/probe-event.c +15 −0
tools/perf/util/probe-finder.h +32 −27

arch/Kconfig+1−3

@@ -135,9 +135,7 @@ config HAVE_DEFAULT_NO_SPIN_MUTEXES · 134 unmodified lines
135	135
136	136	config HAVE_HW_BREAKPOINT
137	137	bool
138		− depends on HAVE_PERF_EVENTS
139		− select ANON_INODES
140		− select PERF_EVENTS
	138	+ depends on PERF_EVENTS
141	139
142	140	config HAVE_USER_RETURN_NOTIFIER
143	141	bool

arch/x86/Kconfig+2−0

@@ -50,6 +50,8 @@ config X86 · 49 unmodified lines
50	50	select HAVE_KERNEL_BZIP2
51	51	select HAVE_KERNEL_LZMA
52	52	select HAVE_HW_BREAKPOINT
	53	+ select PERF_EVENTS
	54	+ select ANON_INODES
53	55	select HAVE_ARCH_KMEMCHECK
54	56	select HAVE_USER_RETURN_NOTIFIER
55	57

arch/x86/include/asm/stacktrace.h+24−0

@@ -5,6 +5,29 @@ extern int kstack_depth_to_print; · 4 unmodified lines
5	5
6	6	int x86_is_stack_id(int id, char *name);
7	7
	8	+struct thread_info;
	9	+struct stacktrace_ops;
	10	+
	11	+typedef unsigned long (walk_stack_t)(struct thread_info tinfo,
	12	+ unsigned long *stack,
	13	+ unsigned long bp,
	14	+ const struct stacktrace_ops *ops,
	15	+ void *data,
	16	+ unsigned long *end,
	17	+ int *graph);
	18	+
	19	+extern unsigned long
	20	+print_context_stack(struct thread_info *tinfo,
	21	+ unsigned long *stack, unsigned long bp,
	22	+ const struct stacktrace_ops ops, void data,
	23	+ unsigned long end, int graph);
	24	+
	25	+extern unsigned long
	26	+print_context_stack_bp(struct thread_info *tinfo,
	27	+ unsigned long *stack, unsigned long bp,
	28	+ const struct stacktrace_ops ops, void data,
	29	+ unsigned long end, int graph);
	30	+
8	31	/* Generic stack tracer with callbacks */
9	32
10	33	struct stacktrace_ops {
@@ -14,6 +37,7 @@ struct stacktrace_ops { · 3 unmodified lines
14	37	void (address)(void data, unsigned long address, int reliable);
15	38	/* On negative return stop dumping */
16	39	int (stack)(void data, char *name);
	40	+ walk_stack_t walk_stack;
17	41	};
18	42
19	43	void dump_trace(struct task_struct tsk, struct pt_regs regs,

arch/x86/kernel/cpu/perf_event.c+1−0

@@ -2336,6 +2336,7 @@ static const struct stacktrace_ops backtrace_ops = { · 2335 unmodified lines
2336	2336	.warning_symbol = backtrace_warning_symbol,
2337	2337	.stack = backtrace_stack,
2338	2338	.address = backtrace_address,
	2339	+ .walk_stack = print_context_stack_bp,
2339	2340	};
2340	2341
2341	2342	#include "../dumpstack.h"

arch/x86/kernel/dumpstack.c+29−4

@@ -109,6 +109,30 @@ print_context_stack(struct thread_info *tinfo, · 108 unmodified lines
109	109	}
110	110	return bp;
111	111	}
	112	+EXPORT_SYMBOL_GPL(print_context_stack);
	113	+
	114	+unsigned long
	115	+print_context_stack_bp(struct thread_info *tinfo,
	116	+ unsigned long *stack, unsigned long bp,
	117	+ const struct stacktrace_ops ops, void data,
	118	+ unsigned long end, int graph)
	119	+{
	120	+ struct stack_frame frame = (struct stack_frame )bp;
	121	+ unsigned long *ret_addr = &frame->return_address;
	122	+
	123	+ while (valid_stack_ptr(tinfo, ret_addr, sizeof(*ret_addr), end)) {
	124	+ unsigned long addr = *ret_addr;
	125	+
	126	+ if (__kernel_text_address(addr)) {
	127	+ ops->address(data, addr, 1);
	128	+ frame = frame->next_frame;
	129	+ ret_addr = &frame->return_address;
	130	+ print_ftrace_graph_addr(addr, data, ops, tinfo, graph);
	131	+ }
	132	+ }
	133	+ return (unsigned long)frame;
	134	+}
	135	+EXPORT_SYMBOL_GPL(print_context_stack_bp);
112	136
113	137
114	138	static void
@@ -141,10 +165,11 @@ static void print_trace_address(void *data, unsigned long addr, int reliable) · 26 unmodified lines
141	165	}
142	166
143	167	static const struct stacktrace_ops print_trace_ops = {
144		− .warning = print_trace_warning,
145		− .warning_symbol = print_trace_warning_symbol,
146		− .stack = print_trace_stack,
147		− .address = print_trace_address,
	168	+ .warning = print_trace_warning,
	169	+ .warning_symbol = print_trace_warning_symbol,
	170	+ .stack = print_trace_stack,
	171	+ .address = print_trace_address,
	172	+ .walk_stack = print_context_stack,
148	173	};
149	174
150	175	void

arch/x86/kernel/dumpstack.h+0−6

@@ -14,12 +14,6 @@ · 13 unmodified lines
14	14	#define get_bp(bp) asm("movq %%rbp, %0" : "=r" (bp) :)
15	15	#endif
16	16
17		−extern unsigned long
18		−print_context_stack(struct thread_info *tinfo,
19		− unsigned long *stack, unsigned long bp,
20		− const struct stacktrace_ops ops, void data,
21		− unsigned long end, int graph);
22		−
23	17	extern void
24	18	show_trace_log_lvl(struct task_struct task, struct pt_regs regs,
25	19	unsigned long stack, unsigned long bp, char log_lvl);

arch/x86/kernel/dumpstack_32.c+1−1

@@ -58,7 +58,7 @@ void dump_trace(struct task_struct task, struct pt_regs regs, · 57 unmodified lines
58	58
59	59	context = (struct thread_info *)
60	60	((unsigned long)stack & (~(THREAD_SIZE - 1)));
61		− bp = print_context_stack(context, stack, bp, ops, data, NULL, &graph);
	61	+ bp = ops->walk_stack(context, stack, bp, ops, data, NULL, &graph);
62	62
63	63	stack = (unsigned long *)context->previous_esp;
64	64	if (!stack)

arch/x86/kernel/dumpstack_64.c+2−2

@@ -188,8 +188,8 @@ void dump_trace(struct task_struct task, struct pt_regs regs, · 187 unmodified lines
188	188	if (ops->stack(data, id) < 0)
189	189	break;
190	190
191		− bp = print_context_stack(tinfo, stack, bp, ops,
192		− data, estack_end, &graph);
	191	+ bp = ops->walk_stack(tinfo, stack, bp, ops,
	192	+ data, estack_end, &graph);
193	193	ops->stack(data, "<EOE>");
194	194	/*
195	195	* We link to the next stack via the

arch/x86/kernel/stacktrace.c+10−8

@@ -53,17 +53,19 @@ save_stack_address_nosched(void *data, unsigned long addr, int reliable) · 52 unmodified lines
53	53	}
54	54
55	55	static const struct stacktrace_ops save_stack_ops = {
56		− .warning = save_stack_warning,
57		− .warning_symbol = save_stack_warning_symbol,
58		− .stack = save_stack_stack,
59		− .address = save_stack_address,
	56	+ .warning = save_stack_warning,
	57	+ .warning_symbol = save_stack_warning_symbol,
	58	+ .stack = save_stack_stack,
	59	+ .address = save_stack_address,
	60	+ .walk_stack = print_context_stack,
60	61	};
61	62
62	63	static const struct stacktrace_ops save_stack_ops_nosched = {
63		− .warning = save_stack_warning,
64		− .warning_symbol = save_stack_warning_symbol,
65		− .stack = save_stack_stack,
66		− .address = save_stack_address_nosched,
	64	+ .warning = save_stack_warning,
	65	+ .warning_symbol = save_stack_warning_symbol,
	66	+ .stack = save_stack_stack,
	67	+ .address = save_stack_address_nosched,
	68	+ .walk_stack = print_context_stack,
67	69	};
68	70
69	71	/*

arch/x86/oprofile/backtrace.c+5−4

@@ -41,10 +41,11 @@ static void backtrace_address(void *data, unsigned long addr, int reliable) · 40 unmodified lines
41	41	}
42	42
43	43	static struct stacktrace_ops backtrace_ops = {
44		− .warning = backtrace_warning,
45		− .warning_symbol = backtrace_warning_symbol,
46		− .stack = backtrace_stack,
47		− .address = backtrace_address,
	44	+ .warning = backtrace_warning,
	45	+ .warning_symbol = backtrace_warning_symbol,
	46	+ .stack = backtrace_stack,
	47	+ .address = backtrace_address,
	48	+ .walk_stack = print_context_stack,
48	49	};
49	50
50	51	struct frame_head {

include/linux/perf_counter.h+0−444

@@ -1,444 +0,0 @@
1		−/*
2		− * NOTE: this file will be removed in a future kernel release, it is
3		− * provided as a courtesy copy of user-space code that relies on the
4		− * old (pre-rename) symbols and constants.
5		− *
6		− * Performance events:
7		− *
8		− * Copyright (C) 2008-2009, Thomas Gleixner <tglx@linutronix.de>
9		− * Copyright (C) 2008-2009, Red Hat, Inc., Ingo Molnar
10		− * Copyright (C) 2008-2009, Red Hat, Inc., Peter Zijlstra
11		− *
12		− * Data type definitions, declarations, prototypes.
13		− *
14		− * Started by: Thomas Gleixner and Ingo Molnar
15		− *
16		− * For licencing details see kernel-base/COPYING
17		− */
18		−#ifndef _LINUX_PERF_COUNTER_H
19		−#define _LINUX_PERF_COUNTER_H
20		−
21		−#include <linux/types.h>
22		−#include <linux/ioctl.h>
23		−#include <asm/byteorder.h>
24		−
25		−/*
26		− * User-space ABI bits:
27		− */
28		−
29		−/*
30		− * attr.type
31		− */
32		−enum perf_type_id {
33		− PERF_TYPE_HARDWARE = 0,
34		− PERF_TYPE_SOFTWARE = 1,
35		− PERF_TYPE_TRACEPOINT = 2,
36		− PERF_TYPE_HW_CACHE = 3,
37		− PERF_TYPE_RAW = 4,
38		−
39		− PERF_TYPE_MAX, /* non-ABI */
40		−};
41		−
42		−/*
43		− * Generalized performance counter event types, used by the
44		− * attr.event_id parameter of the sys_perf_counter_open()
45		− * syscall:
46		− */
47		−enum perf_hw_id {
48		− /*
49		− * Common hardware events, generalized by the kernel:
50		− */
51		− PERF_COUNT_HW_CPU_CYCLES = 0,
52		− PERF_COUNT_HW_INSTRUCTIONS = 1,
53		− PERF_COUNT_HW_CACHE_REFERENCES = 2,
54		− PERF_COUNT_HW_CACHE_MISSES = 3,
55		− PERF_COUNT_HW_BRANCH_INSTRUCTIONS = 4,
56		− PERF_COUNT_HW_BRANCH_MISSES = 5,
57		− PERF_COUNT_HW_BUS_CYCLES = 6,
58		−
59		− PERF_COUNT_HW_MAX, /* non-ABI */
60		−};
61		−
62		−/*
63		− * Generalized hardware cache counters:
64		− *
65		− * { L1-D, L1-I, LLC, ITLB, DTLB, BPU } x
66		− * { read, write, prefetch } x
67		− * { accesses, misses }
68		− */
69		−enum perf_hw_cache_id {
70		− PERF_COUNT_HW_CACHE_L1D = 0,
71		− PERF_COUNT_HW_CACHE_L1I = 1,
72		− PERF_COUNT_HW_CACHE_LL = 2,
73		− PERF_COUNT_HW_CACHE_DTLB = 3,
74		− PERF_COUNT_HW_CACHE_ITLB = 4,
75		− PERF_COUNT_HW_CACHE_BPU = 5,
76		−
77		− PERF_COUNT_HW_CACHE_MAX, /* non-ABI */
78		−};
79		−
80		−enum perf_hw_cache_op_id {
81		− PERF_COUNT_HW_CACHE_OP_READ = 0,
82		− PERF_COUNT_HW_CACHE_OP_WRITE = 1,
83		− PERF_COUNT_HW_CACHE_OP_PREFETCH = 2,
84		−
85		− PERF_COUNT_HW_CACHE_OP_MAX, /* non-ABI */
86		−};
87		−
88		−enum perf_hw_cache_op_result_id {
89		− PERF_COUNT_HW_CACHE_RESULT_ACCESS = 0,
90		− PERF_COUNT_HW_CACHE_RESULT_MISS = 1,
91		−
92		− PERF_COUNT_HW_CACHE_RESULT_MAX, /* non-ABI */
93		−};
94		−
95		−/*
96		− * Special "software" counters provided by the kernel, even if the hardware
97		− * does not support performance counters. These counters measure various
98		− * physical and sw events of the kernel (and allow the profiling of them as
99		− * well):
100		− */
101		−enum perf_sw_ids {
102		− PERF_COUNT_SW_CPU_CLOCK = 0,
103		− PERF_COUNT_SW_TASK_CLOCK = 1,
104		− PERF_COUNT_SW_PAGE_FAULTS = 2,
105		− PERF_COUNT_SW_CONTEXT_SWITCHES = 3,
106		− PERF_COUNT_SW_CPU_MIGRATIONS = 4,
107		− PERF_COUNT_SW_PAGE_FAULTS_MIN = 5,
108		− PERF_COUNT_SW_PAGE_FAULTS_MAJ = 6,
109		− PERF_COUNT_SW_ALIGNMENT_FAULTS = 7,
110		− PERF_COUNT_SW_EMULATION_FAULTS = 8,
111		−
112		− PERF_COUNT_SW_MAX, /* non-ABI */
113		−};
114		−
115		−/*
116		− * Bits that can be set in attr.sample_type to request information
117		− * in the overflow packets.
118		− */
119		−enum perf_counter_sample_format {
120		− PERF_SAMPLE_IP = 1U << 0,
121		− PERF_SAMPLE_TID = 1U << 1,
122		− PERF_SAMPLE_TIME = 1U << 2,
123		− PERF_SAMPLE_ADDR = 1U << 3,
124		− PERF_SAMPLE_READ = 1U << 4,
125		− PERF_SAMPLE_CALLCHAIN = 1U << 5,
126		− PERF_SAMPLE_ID = 1U << 6,
127		− PERF_SAMPLE_CPU = 1U << 7,
128		− PERF_SAMPLE_PERIOD = 1U << 8,
129		− PERF_SAMPLE_STREAM_ID = 1U << 9,
130		− PERF_SAMPLE_RAW = 1U << 10,
131		−
132		− PERF_SAMPLE_MAX = 1U << 11, /* non-ABI */
133		−};
134		−
135		−/*
136		− * The format of the data returned by read() on a perf counter fd,
137		− * as specified by attr.read_format:
138		− *
139		− * struct read_format {
140		− * { u64 value;
141		− * { u64 time_enabled; } && PERF_FORMAT_ENABLED
142		− * { u64 time_running; } && PERF_FORMAT_RUNNING
143		− * { u64 id; } && PERF_FORMAT_ID
144		− * } && !PERF_FORMAT_GROUP
145		− *
146		− * { u64 nr;
147		− * { u64 time_enabled; } && PERF_FORMAT_ENABLED
148		− * { u64 time_running; } && PERF_FORMAT_RUNNING
149		− * { u64 value;
150		− * { u64 id; } && PERF_FORMAT_ID
151		− * } cntr[nr];
152		− * } && PERF_FORMAT_GROUP
153		− * };
154		− */
155		−enum perf_counter_read_format {
156		− PERF_FORMAT_TOTAL_TIME_ENABLED = 1U << 0,
157		− PERF_FORMAT_TOTAL_TIME_RUNNING = 1U << 1,
158		− PERF_FORMAT_ID = 1U << 2,
159		− PERF_FORMAT_GROUP = 1U << 3,
160		−
161		− PERF_FORMAT_MAX = 1U << 4, /* non-ABI */
162		−};
163		−
164		−#define PERF_ATTR_SIZE_VER0 64 /* sizeof first published struct */
165		−
166		−/*
167		− * Hardware event to monitor via a performance monitoring counter:
168		− */
169		−struct perf_counter_attr {
170		−
171		− /*
172		− * Major type: hardware/software/tracepoint/etc.
173		− */
174		− __u32 type;
175		−
176		− /*
177		− * Size of the attr structure, for fwd/bwd compat.
178		− */
179		− __u32 size;
180		−
181		− /*
182		− * Type specific configuration information.
183		− */
184		− __u64 config;
185		−
186		− union {
187		− __u64 sample_period;
188		− __u64 sample_freq;
189		− };
190		−
191		− __u64 sample_type;
192		− __u64 read_format;
193		−
194		− __u64 disabled : 1, /* off by default */
195		− inherit : 1, /* children inherit it */
196		− pinned : 1, /* must always be on PMU */
197		− exclusive : 1, /* only group on PMU */
198		− exclude_user : 1, /* don't count user */
199		− exclude_kernel : 1, /* ditto kernel */
200		− exclude_hv : 1, /* ditto hypervisor */
201		− exclude_idle : 1, /* don't count when idle */
202		− mmap : 1, /* include mmap data */
203		− comm : 1, /* include comm data */
204		− freq : 1, /* use freq, not period */
205		− inherit_stat : 1, /* per task counts */
206		− enable_on_exec : 1, /* next exec enables */
207		− task : 1, /* trace fork/exit */
208		− watermark : 1, /* wakeup_watermark */
209		−
210		− __reserved_1 : 49;
211		−
212		− union {
213		− __u32 wakeup_events; /* wakeup every n events */
214		− __u32 wakeup_watermark; /* bytes before wakeup */
215		− };
216		− __u32 __reserved_2;
217		−
218		− __u64 __reserved_3;
219		−};
220		−
221		−/*
222		− * Ioctls that can be done on a perf counter fd:
223		− */
224		−#define PERF_COUNTER_IOC_ENABLE _IO ('$', 0)
225		−#define PERF_COUNTER_IOC_DISABLE _IO ('$', 1)
226		−#define PERF_COUNTER_IOC_REFRESH _IO ('$', 2)
227		−#define PERF_COUNTER_IOC_RESET _IO ('$', 3)
228		−#define PERF_COUNTER_IOC_PERIOD _IOW('$', 4, u64)
229		−#define PERF_COUNTER_IOC_SET_OUTPUT _IO ('$', 5)
230		−#define PERF_COUNTER_IOC_SET_FILTER _IOW('$', 6, char *)
231		−
232		−enum perf_counter_ioc_flags {
233		− PERF_IOC_FLAG_GROUP = 1U << 0,
234		−};
235		−
236		−/*
237		− * Structure of the page that can be mapped via mmap
238		− */
239		−struct perf_counter_mmap_page {
240		− __u32 version; /* version number of this structure */
241		− __u32 compat_version; /* lowest version this is compat with */
242		−
243		− /*
244		− * Bits needed to read the hw counters in user-space.
245		− *
246		− * u32 seq;
247		− * s64 count;
248		− *
249		− * do {
250		− * seq = pc->lock;
251		− *
252		− * barrier()
253		− * if (pc->index) {
254		− * count = pmc_read(pc->index - 1);
255		− * count += pc->offset;
256		− * } else
257		− * goto regular_read;
258		− *
259		− * barrier();
260		− * } while (pc->lock != seq);
261		− *
262		− * NOTE: for obvious reason this only works on self-monitoring
263		− * processes.
264		− */
265		− __u32 lock; /* seqlock for synchronization */
266		− __u32 index; /* hardware counter identifier */
267		− __s64 offset; /* add to hardware counter value */
268		− __u64 time_enabled; /* time counter active */
269		− __u64 time_running; /* time counter on cpu */
270		−
271		− /*
272		− * Hole for extension of the self monitor capabilities
273		− */
274		−
275		− __u64 __reserved[123]; /* align to 1k */
276		−
277		− /*
278		− * Control data for the mmap() data buffer.
279		− *
280		− * User-space reading the @data_head value should issue an rmb(), on
281		− * SMP capable platforms, after reading this value -- see
282		− * perf_counter_wakeup().
283		− *
284		− * When the mapping is PROT_WRITE the @data_tail value should be
285		− * written by userspace to reflect the last read data. In this case
286		− * the kernel will not over-write unread data.
287		− */
288		− __u64 data_head; /* head in the data section */
289		− __u64 data_tail; /* user-space written tail */
290		−};
291		−
292		−#define PERF_EVENT_MISC_CPUMODE_MASK (3 << 0)
293		−#define PERF_EVENT_MISC_CPUMODE_UNKNOWN (0 << 0)
294		−#define PERF_EVENT_MISC_KERNEL (1 << 0)
295		−#define PERF_EVENT_MISC_USER (2 << 0)
296		−#define PERF_EVENT_MISC_HYPERVISOR (3 << 0)
297		−
298		−struct perf_event_header {
299		− __u32 type;
300		− __u16 misc;
301		− __u16 size;
302		−};
303		−
304		−enum perf_event_type {
305		−
306		− /*
307		− * The MMAP events record the PROT_EXEC mappings so that we can
308		− * correlate userspace IPs to code. They have the following structure:
309		− *
310		− * struct {
311		− * struct perf_event_header header;
312		− *
313		− * u32 pid, tid;
314		− * u64 addr;
315		− * u64 len;
316		− * u64 pgoff;
317		− * char filename[];
318		− * };
319		− */
320		− PERF_EVENT_MMAP = 1,
321		−
322		− /*
323		− * struct {
324		− * struct perf_event_header header;
325		− * u64 id;
326		− * u64 lost;
327		− * };
328		− */
329		− PERF_EVENT_LOST = 2,
330		−
331		− /*
332		− * struct {
333		− * struct perf_event_header header;
334		− *
335		− * u32 pid, tid;
336		− * char comm[];
337		− * };
338		− */
339		− PERF_EVENT_COMM = 3,
340		−
341		− /*
342		− * struct {
343		− * struct perf_event_header header;
344		− * u32 pid, ppid;
345		− * u32 tid, ptid;
346		− * u64 time;
347		− * };
348		− */
349		− PERF_EVENT_EXIT = 4,
350		−
351		− /*
352		− * struct {
353		− * struct perf_event_header header;
354		− * u64 time;
355		− * u64 id;
356		− * u64 stream_id;
357		− * };
358		− */
359		− PERF_EVENT_THROTTLE = 5,
360		− PERF_EVENT_UNTHROTTLE = 6,
361		−
362		− /*
363		− * struct {
364		− * struct perf_event_header header;
365		− * u32 pid, ppid;
366		− * u32 tid, ptid;
367		− * u64 time;
368		− * };
369		− */
370		− PERF_EVENT_FORK = 7,
371		−
372		− /*
373		− * struct {
374		− * struct perf_event_header header;
375		− * u32 pid, tid;
376		− *
377		− * struct read_format values;
378		− * };
379		− */
380		− PERF_EVENT_READ = 8,
381		−
382		− /*
383		− * struct {
384		− * struct perf_event_header header;
385		− *
386		− * { u64 ip; } && PERF_SAMPLE_IP
387		− * { u32 pid, tid; } && PERF_SAMPLE_TID
388		− * { u64 time; } && PERF_SAMPLE_TIME
389		− * { u64 addr; } && PERF_SAMPLE_ADDR
390		− * { u64 id; } && PERF_SAMPLE_ID
391		− * { u64 stream_id;} && PERF_SAMPLE_STREAM_ID
392		− * { u32 cpu, res; } && PERF_SAMPLE_CPU
393		− * { u64 period; } && PERF_SAMPLE_PERIOD
394		− *
395		− * { struct read_format values; } && PERF_SAMPLE_READ
396		− *
397		− * { u64 nr,
398		− * u64 ips[nr]; } && PERF_SAMPLE_CALLCHAIN
399		− *
400		− * #
401		− * # The RAW record below is opaque data wrt the ABI
402		− * #
403		− * # That is, the ABI doesn't make any promises wrt to
404		− * # the stability of its content, it may vary depending
405		− * # on event, hardware, kernel version and phase of
406		− * # the moon.
407		− * #
408		− * # In other words, PERF_SAMPLE_RAW contents are not an ABI.
409		− * #
410		− *
411		− * { u32 size;
412		− * char data[size];}&& PERF_SAMPLE_RAW
413		− * };
414		− */
415		− PERF_EVENT_SAMPLE = 9,
416		−
417		− PERF_EVENT_MAX, /* non-ABI */
418		−};
419		−
420		−enum perf_callchain_context {
421		− PERF_CONTEXT_HV = (__u64)-32,
422		− PERF_CONTEXT_KERNEL = (__u64)-128,
423		− PERF_CONTEXT_USER = (__u64)-512,
424		−
425		− PERF_CONTEXT_GUEST = (__u64)-2048,
426		− PERF_CONTEXT_GUEST_KERNEL = (__u64)-2176,
427		− PERF_CONTEXT_GUEST_USER = (__u64)-2560,
428		−
429		− PERF_CONTEXT_MAX = (__u64)-4095,
430		−};
431		−
432		−#define PERF_FLAG_FD_NO_GROUP (1U << 0)
433		−#define PERF_FLAG_FD_OUTPUT (1U << 1)
434		−
435		−/*
436		− * In case some app still references the old symbols:
437		− */
438		−
439		−#define __NR_perf_counter_open __NR_perf_event_open
440		−
441		−#define PR_TASK_PERF_COUNTERS_DISABLE PR_TASK_PERF_EVENTS_DISABLE
442		−#define PR_TASK_PERF_COUNTERS_ENABLE PR_TASK_PERF_EVENTS_ENABLE
443		−
444		−#endif /* _LINUX_PERF_COUNTER_H */

kernel/perf_event.c+18−14

@@ -1381,6 +1381,9 @@ static void perf_ctx_adjust_freq(struct perf_event_context *ctx) · 1380 unmodified lines
1381	1381	if (event->state != PERF_EVENT_STATE_ACTIVE)
1382	1382	continue;
1383	1383
	1384	+ if (event->cpu != -1 && event->cpu != smp_processor_id())
	1385	+ continue;
	1386	+
1384	1387	hwc = &event->hw;
1385	1388
1386	1389	interrupts = hwc->interrupts;
@@ -3265,6 +3268,9 @@ static void perf_event_task_output(struct perf_event *event, · 1878 unmodified lines
3265	3268
3266	3269	static int perf_event_task_match(struct perf_event *event)
3267	3270	{
	3271	+ if (event->cpu != -1 && event->cpu != smp_processor_id())
	3272	+ return 0;
	3273	+
3268	3274	if (event->attr.comm \|\| event->attr.mmap \|\| event->attr.task)
3269	3275	return 1;
3270	3276
@@ -3290,12 +3296,11 @@ static void perf_event_task_event(struct perf_task_event *task_event) · 19 unmodified lines
3290	3296	rcu_read_lock();
3291	3297	cpuctx = &get_cpu_var(perf_cpu_context);
3292	3298	perf_event_task_ctx(&cpuctx->ctx, task_event);
3293		− put_cpu_var(perf_cpu_context);
3294		−
3295	3299	if (!ctx)
3296	3300	ctx = rcu_dereference(task_event->task->perf_event_ctxp);
3297	3301	if (ctx)
3298	3302	perf_event_task_ctx(ctx, task_event);
	3303	+ put_cpu_var(perf_cpu_context);
3299	3304	rcu_read_unlock();
3300	3305	}
3301	3306
@@ -3372,6 +3377,9 @@ static void perf_event_comm_output(struct perf_event *event, · 70 unmodified lines
3372	3377
3373	3378	static int perf_event_comm_match(struct perf_event *event)
3374	3379	{
	3380	+ if (event->cpu != -1 && event->cpu != smp_processor_id())
	3381	+ return 0;
	3382	+
3375	3383	if (event->attr.comm)
3376	3384	return 1;
3377	3385
@@ -3408,15 +3416,10 @@ static void perf_event_comm_event(struct perf_comm_event *comm_event) · 30 unmodified lines
3408	3416	rcu_read_lock();
3409	3417	cpuctx = &get_cpu_var(perf_cpu_context);
3410	3418	perf_event_comm_ctx(&cpuctx->ctx, comm_event);
3411		− put_cpu_var(perf_cpu_context);
3412		−
3413		− /*
3414		− * doesn't really matter which of the child contexts the
3415		− * events ends up in.
3416		− */
3417	3419	ctx = rcu_dereference(current->perf_event_ctxp);
3418	3420	if (ctx)
3419	3421	perf_event_comm_ctx(ctx, comm_event);
	3422	+ put_cpu_var(perf_cpu_context);
3420	3423	rcu_read_unlock();
3421	3424	}
3422	3425
@@ -3491,6 +3494,9 @@ static void perf_event_mmap_output(struct perf_event *event, · 68 unmodified lines
3491	3494	static int perf_event_mmap_match(struct perf_event *event,
3492	3495	struct perf_mmap_event *mmap_event)
3493	3496	{
	3497	+ if (event->cpu != -1 && event->cpu != smp_processor_id())
	3498	+ return 0;
	3499	+
3494	3500	if (event->attr.mmap)
3495	3501	return 1;
3496	3502
@@ -3564,15 +3570,10 @@ got_name: · 67 unmodified lines
3564	3570	rcu_read_lock();
3565	3571	cpuctx = &get_cpu_var(perf_cpu_context);
3566	3572	perf_event_mmap_ctx(&cpuctx->ctx, mmap_event);
3567		− put_cpu_var(perf_cpu_context);
3568		−
3569		− /*
3570		− * doesn't really matter which of the child contexts the
3571		− * events ends up in.
3572		− */
3573	3573	ctx = rcu_dereference(current->perf_event_ctxp);
3574	3574	if (ctx)
3575	3575	perf_event_mmap_ctx(ctx, mmap_event);
	3576	+ put_cpu_var(perf_cpu_context);
3576	3577	rcu_read_unlock();
3577	3578
3578	3579	kfree(buf);
@@ -3863,6 +3864,9 @@ static int perf_swevent_match(struct perf_event *event, · 284 unmodified lines
3863	3864	struct perf_sample_data *data,
3864	3865	struct pt_regs *regs)
3865	3866	{
	3867	+ if (event->cpu != -1 && event->cpu != smp_processor_id())
	3868	+ return 0;
	3869	+
3866	3870	if (!perf_swevent_is_counting(event))
3867	3871	return 0;
3868	3872

kernel/trace/trace_kprobe.c+25−6

@@ -282,6 +282,18 @@ static int kprobe_dispatcher(struct kprobe kp, struct pt_regs regs); · 281 unmodified lines
282	282	static int kretprobe_dispatcher(struct kretprobe_instance *ri,
283	283	struct pt_regs *regs);
284	284
	285	+/* Check the name is good for event/group */
	286	+static int check_event_name(const char *name)
	287	+{
	288	+ if (!isalpha(name) && name != '_')
	289	+ return 0;
	290	+ while (*++name != '\0') {
	291	+ if (!isalpha(name) && !isdigit(name) && *name != '_')
	292	+ return 0;
	293	+ }
	294	+ return 1;
	295	+}
	296	+
285	297	/*
286	298	* Allocate new trace_probe and initialize it (including kprobes).
287	299	*/
@@ -293,10 +305,11 @@ static struct trace_probe alloc_trace_probe(const char group, · 5 unmodified lines
293	305	int nargs, int is_return)
294	306	{
295	307	struct trace_probe *tp;
	308	+ int ret = -ENOMEM;
296	309
297	310	tp = kzalloc(SIZEOF_TRACE_PROBE(nargs), GFP_KERNEL);
298	311	if (!tp)
299		− return ERR_PTR(-ENOMEM);
	312	+ return ERR_PTR(ret);
300	313
301	314	if (symbol) {
302	315	tp->symbol = kstrdup(symbol, GFP_KERNEL);
@@ -312,14 +325,20 @@ static struct trace_probe alloc_trace_probe(const char group, · 9 unmodified lines
312	325	else
313	326	tp->rp.kp.pre_handler = kprobe_dispatcher;
314	327
315		− if (!event)
	328	+ if (!event \|\| !check_event_name(event)) {
	329	+ ret = -EINVAL;
316	330	goto error;
	331	+ }
	332	+
317	333	tp->call.name = kstrdup(event, GFP_KERNEL);
318	334	if (!tp->call.name)
319	335	goto error;
320	336
321		− if (!group)
	337	+ if (!group \|\| !check_event_name(group)) {
	338	+ ret = -EINVAL;
322	339	goto error;
	340	+ }
	341	+
323	342	tp->call.system = kstrdup(group, GFP_KERNEL);
324	343	if (!tp->call.system)
325	344	goto error;
@@ -330,7 +349,7 @@ error: · 4 unmodified lines
330	349	kfree(tp->call.name);
331	350	kfree(tp->symbol);
332	351	kfree(tp);
333		− return ERR_PTR(-ENOMEM);
	352	+ return ERR_PTR(ret);
334	353	}
335	354
336	355	static void free_probe_arg(struct probe_arg *arg)
@@ -695,10 +714,10 @@ static int create_trace_probe(int argc, char **argv) · 358 unmodified lines
695	714	if (!event) {
696	715	/* Make a new event name */
697	716	if (symbol)
698		− snprintf(buf, MAX_EVENT_NAME_LEN, "%c@%s%+ld",
	717	+ snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
699	718	is_return ? 'r' : 'p', symbol, offset);
700	719	else
701		− snprintf(buf, MAX_EVENT_NAME_LEN, "%c@0x%p",
	720	+ snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
702	721	is_return ? 'r' : 'p', addr);
703	722	event = buf;
704	723	}

kernel/trace/trace_sysprof.c+1−0

@@ -93,6 +93,7 @@ static const struct stacktrace_ops backtrace_ops = { · 92 unmodified lines
93	93	.warning_symbol = backtrace_warning_symbol,
94	94	.stack = backtrace_stack,
95	95	.address = backtrace_address,
	96	+ .walk_stack = print_context_stack,
96	97	};
97	98
98	99	static int

tools/perf/Makefile+2−1

@@ -487,10 +487,11 @@ else · 486 unmodified lines
487	487	msg := $(error No libelf.h/libelf found, please install libelf-dev/elfutils-libelf-devel and glibc-dev[el]);
488	488	endif
489	489
490		−ifneq ($(shell sh -c "(echo '\#include <libdwarf/dwarf.h>'; echo '\#include <libdwarf/libdwarf.h>'; echo 'int main(void) { Dwarf_Debug dbg; Dwarf_Error err; Dwarf_Ranges *rng; dwarf_init(0, DW_DLC_READ, 0, 0, &dbg, &err); dwarf_get_ranges(dbg, 0, &rng, 0, 0, &err); return (long)dbg; }') \| $(CC) -x c - $(ALL_CFLAGS) -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -ldwarf -lelf -o /dev/null $(ALL_LDFLAGS) $(EXTLIBS) "$(QUIET_STDERR)" && echo y"), y)
	490	+ifneq ($(shell sh -c "(echo '\#ifndef _MIPS_SZLONG'; echo '\#define _MIPS_SZLONG 0'; echo '\#endif'; echo '\#include <dwarf.h>'; echo '\#include <libdwarf.h>'; echo 'int main(void) { Dwarf_Debug dbg; Dwarf_Error err; Dwarf_Ranges *rng; dwarf_init(0, DW_DLC_READ, 0, 0, &dbg, &err); dwarf_get_ranges(dbg, 0, &rng, 0, 0, &err); return (long)dbg; }') \| $(CC) -x c - $(ALL_CFLAGS) -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -I/usr/include/libdwarf -ldwarf -lelf -o /dev/null $(ALL_LDFLAGS) $(EXTLIBS) "$(QUIET_STDERR)" && echo y"), y)
491	491	msg := $(warning No libdwarf.h found or old libdwarf.h found, disables dwarf support. Please install libdwarf-dev/libdwarf-devel >= 20081231);
492	492	BASIC_CFLAGS += -DNO_LIBDWARF
493	493	else
	494	+ BASIC_CFLAGS += -I/usr/include/libdwarf
494	495	EXTLIBS += -lelf -ldwarf
495	496	LIB_OBJS += util/probe-finder.o
496	497	endif

tools/perf/builtin-probe.c+4−0

@@ -38,6 +38,7 @@ · 37 unmodified lines
38	38	#include "util/strlist.h"
39	39	#include "util/event.h"
40	40	#include "util/debug.h"
	41	+#include "util/debugfs.h"
41	42	#include "util/symbol.h"
42	43	#include "util/thread.h"
43	44	#include "util/session.h"
@@ -205,6 +206,9 @@ int cmd_probe(int argc, const char *argv, const char prefix __used) · 161 unmodified lines
205	206	if ((!session.nr_probe && !session.dellist && !session.list_events))
206	207	usage_with_options(probe_usage, options);
207	208
	209	+ if (debugfs_valid_mountpoint(debugfs_path) < 0)
	210	+ die("Failed to find debugfs path.");
	211	+
208	212	if (session.list_events) {
209	213	if (session.nr_probe != 0 \|\| session.dellist) {
210	214	pr_warning(" Error: Don't use --list with"

tools/perf/builtin-report.c+1−1

@@ -224,7 +224,7 @@ static int __cmd_report(void) · 223 unmodified lines
224	224
225	225	perf_session__collapse_resort(session);
226	226	perf_session__output_resort(session, session->events_stats.total);
227		− fprintf(stdout, "# Samples: %ld\n#\n", session->events_stats.total);
	227	+ fprintf(stdout, "# Samples: %Ld\n#\n", session->events_stats.total);
228	228	perf_session__fprintf_hists(session, NULL, false, stdout);
229	229	if (sort_order == default_sort_order &&
230	230	parent_pattern == default_parent_pattern)

tools/perf/util/event.h+2−2

@@ -95,8 +95,8 @@ typedef union event_union { · 94 unmodified lines
95	95	} event_t;
96	96
97	97	struct events_stats {
98		− unsigned long total;
99		− unsigned long lost;
	98	+ u64 total;
	99	+ u64 lost;
100	100	};
101	101
102	102	void event__print_totals(void);

tools/perf/util/probe-event.c+15−0

@@ -62,6 +62,18 @@ static int e_snprintf(char str, size_t size, const char format, ...) · 61 unmodified lines
62	62	return ret;
63	63	}
64	64
	65	+/* Check the name is good for event/group */
	66	+static bool check_event_name(const char *name)
	67	+{
	68	+ if (!isalpha(name) && name != '_')
	69	+ return false;
	70	+ while (*++name != '\0') {
	71	+ if (!isalpha(name) && !isdigit(name) && *name != '_')
	72	+ return false;
	73	+ }
	74	+ return true;
	75	+}
	76	+
65	77	/* Parse probepoint definition. */
66	78	static void parse_perf_probe_probepoint(char arg, struct probe_point pp)
67	79	{
@@ -82,6 +94,9 @@ static void parse_perf_probe_probepoint(char arg, struct probe_point pp) · 14 unmodified lines
82	94	ptr = strchr(arg, ':');
83	95	if (ptr) /* Group name is not supported yet. */
84	96	semantic_error("Group name is not supported yet.");
	97	+ if (!check_event_name(arg))
	98	+ semantic_error("%s is bad for event name -it must "
	99	+ "follow C symbol-naming rule.", arg);
85	100	pp->event = strdup(arg);
86	101	arg = tmp;
87	102	}

tools/perf/util/probe-finder.h+32−27

@@ -1,9 +1,9 @@
1	1	#ifndef _PROBE_FINDER_H
2	2	#define _PROBE_FINDER_H
3	3
4		−#define MAX_PATH_LEN 256
5		−#define MAX_PROBE_BUFFER 1024
6		−#define MAX_PROBES 128
	4	+#define MAX_PATH_LEN 256
	5	+#define MAX_PROBE_BUFFER 1024
	6	+#define MAX_PROBES 128
7	7
8	8	static inline int is_c_varname(const char *name)
9	9	{
@@ -12,48 +12,53 @@ static inline int is_c_varname(const char *name) · 2 unmodified lines
12	12	}
13	13
14	14	struct probe_point {
15		− char event; / Event name */
16		− char group; / Event group */
	15	+ char event; / Event name */
	16	+ char group; / Event group */
17	17
18	18	/* Inputs */
19		− char file; / File name */
20		− int line; /* Line number */
	19	+ char file; / File name */
	20	+ int line; /* Line number */
21	21
22		− char function; / Function name */
23		− int offset; /* Offset bytes */
	22	+ char function; / Function name */
	23	+ int offset; /* Offset bytes */
24	24
25		− int nr_args; /* Number of arguments */
26		− char *args; / Arguments */
	25	+ int nr_args; /* Number of arguments */
	26	+ char *args; / Arguments */
27	27
28		− int retprobe; /* Return probe */
	28	+ int retprobe; /* Return probe */
29	29
30	30	/* Output */
31		− int found; /* Number of found probe points */
32		− char probes[MAX_PROBES]; / Output buffers (will be allocated)*/
	31	+ int found; /* Number of found probe points */
	32	+ char probes[MAX_PROBES]; / Output buffers (will be allocated)*/
33	33	};
34	34
35	35	#ifndef NO_LIBDWARF
36	36	extern int find_probepoint(int fd, struct probe_point *pp);
37	37
38		−#include <libdwarf/dwarf.h>
39		−#include <libdwarf/libdwarf.h>
	38	+/* Workaround for undefined _MIPS_SZLONG bug in libdwarf.h: */
	39	+#ifndef _MIPS_SZLONG
	40	+# define _MIPS_SZLONG 0
	41	+#endif
	42	+
	43	+#include <dwarf.h>
	44	+#include <libdwarf.h>
40	45
41	46	struct probe_finder {
42		− struct probe_point pp; / Target probe point */
	47	+ struct probe_point pp; / Target probe point */
43	48
44	49	/* For function searching */
45		− Dwarf_Addr addr; /* Address */
46		− Dwarf_Unsigned fno; /* File number */
47		− Dwarf_Unsigned lno; /* Line number */
48		− Dwarf_Off inl_offs; /* Inline offset */
49		− Dwarf_Die cu_die; /* Current CU */
	50	+ Dwarf_Addr addr; /* Address */
	51	+ Dwarf_Unsigned fno; /* File number */
	52	+ Dwarf_Unsigned lno; /* Line number */
	53	+ Dwarf_Off inl_offs; /* Inline offset */
	54	+ Dwarf_Die cu_die; /* Current CU */
50	55
51	56	/* For variable searching */
52		− Dwarf_Addr cu_base; /* Current CU base address */
53		− Dwarf_Locdesc fbloc; /* Location of Current Frame Base */
54		− const char var; / Current variable name */
55		− char buf; / Current output buffer */
56		− int len; /* Length of output buffer */
	57	+ Dwarf_Addr cu_base; /* Current CU base address */
	58	+ Dwarf_Locdesc fbloc; /* Location of Current Frame Base */
	59	+ const char var; / Current variable name */
	60	+ char buf; / Current output buffer */
	61	+ int len; /* Length of output buffer */
57	62	};
58	63	#endif /* NO_LIBDWARF */
59	64