aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--arch/powerpc/Kconfig1
-rw-r--r--arch/powerpc/include/asm/hw_irq.h6
-rw-r--r--arch/powerpc/include/asm/perf_counter.h52
-rw-r--r--arch/powerpc/kernel/Makefile8
-rw-r--r--arch/powerpc/kernel/mpc7450-pmu.c417
-rw-r--r--arch/powerpc/kernel/perf_counter.c257
-rw-r--r--arch/powerpc/kernel/power4-pmu.c89
-rw-r--r--arch/powerpc/kernel/power5+-pmu.c95
-rw-r--r--arch/powerpc/kernel/power5-pmu.c98
-rw-r--r--arch/powerpc/kernel/power6-pmu.c72
-rw-r--r--arch/powerpc/kernel/power7-pmu.c61
-rw-r--r--arch/powerpc/kernel/ppc970-pmu.c63
-rw-r--r--arch/powerpc/kernel/time.c25
-rw-r--r--arch/powerpc/platforms/Kconfig.cputype12
-rw-r--r--arch/x86/include/asm/perf_counter.h5
-rw-r--r--arch/x86/include/asm/pgtable_32.h8
-rw-r--r--arch/x86/include/asm/uaccess.h7
-rw-r--r--arch/x86/kernel/cpu/perf_counter.c138
-rw-r--r--arch/x86/mm/gup.c58
-rw-r--r--fs/anon_inodes.c15
-rw-r--r--include/asm-generic/kmap_types.h5
-rw-r--r--include/linux/mm.h6
-rw-r--r--include/linux/perf_counter.h79
-rw-r--r--kernel/perf_counter.c312
-rw-r--r--kernel/sched.c3
-rw-r--r--tools/perf/Makefile10
-rw-r--r--tools/perf/builtin-annotate.c262
-rw-r--r--tools/perf/builtin-record.c163
-rw-r--r--tools/perf/builtin-report.c439
-rw-r--r--tools/perf/builtin-stat.c308
-rw-r--r--tools/perf/builtin-top.c24
-rw-r--r--tools/perf/perf.h7
-rw-r--r--tools/perf/types.h17
-rw-r--r--tools/perf/util/ctype.c17
-rw-r--r--tools/perf/util/parse-events.c14
-rw-r--r--tools/perf/util/string.c2
-rw-r--r--tools/perf/util/string.h4
-rw-r--r--tools/perf/util/symbol.c20
-rw-r--r--tools/perf/util/symbol.h16
-rw-r--r--tools/perf/util/util.h18
40 files changed, 2321 insertions, 892 deletions
diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig
index 9fb344d5a86a..bf6cedfa05db 100644
--- a/arch/powerpc/Kconfig
+++ b/arch/powerpc/Kconfig
@@ -126,6 +126,7 @@ config PPC
126 select HAVE_OPROFILE 126 select HAVE_OPROFILE
127 select HAVE_SYSCALL_WRAPPERS if PPC64 127 select HAVE_SYSCALL_WRAPPERS if PPC64
128 select GENERIC_ATOMIC64 if PPC32 128 select GENERIC_ATOMIC64 if PPC32
129 select HAVE_PERF_COUNTERS
129 130
130config EARLY_PRINTK 131config EARLY_PRINTK
131 bool 132 bool
diff --git a/arch/powerpc/include/asm/hw_irq.h b/arch/powerpc/include/asm/hw_irq.h
index b7f8f4a87cc0..867ab8ed69b3 100644
--- a/arch/powerpc/include/asm/hw_irq.h
+++ b/arch/powerpc/include/asm/hw_irq.h
@@ -131,6 +131,8 @@ static inline int irqs_disabled_flags(unsigned long flags)
131struct irq_chip; 131struct irq_chip;
132 132
133#ifdef CONFIG_PERF_COUNTERS 133#ifdef CONFIG_PERF_COUNTERS
134
135#ifdef CONFIG_PPC64
134static inline unsigned long test_perf_counter_pending(void) 136static inline unsigned long test_perf_counter_pending(void)
135{ 137{
136 unsigned long x; 138 unsigned long x;
@@ -154,15 +156,15 @@ static inline void clear_perf_counter_pending(void)
154 "r" (0), 156 "r" (0),
155 "i" (offsetof(struct paca_struct, perf_counter_pending))); 157 "i" (offsetof(struct paca_struct, perf_counter_pending)));
156} 158}
159#endif /* CONFIG_PPC64 */
157 160
158#else 161#else /* CONFIG_PERF_COUNTERS */
159 162
160static inline unsigned long test_perf_counter_pending(void) 163static inline unsigned long test_perf_counter_pending(void)
161{ 164{
162 return 0; 165 return 0;
163} 166}
164 167
165static inline void set_perf_counter_pending(void) {}
166static inline void clear_perf_counter_pending(void) {} 168static inline void clear_perf_counter_pending(void) {}
167#endif /* CONFIG_PERF_COUNTERS */ 169#endif /* CONFIG_PERF_COUNTERS */
168 170
diff --git a/arch/powerpc/include/asm/perf_counter.h b/arch/powerpc/include/asm/perf_counter.h
index cc7c887705b8..8ccd4e155768 100644
--- a/arch/powerpc/include/asm/perf_counter.h
+++ b/arch/powerpc/include/asm/perf_counter.h
@@ -10,6 +10,8 @@
10 */ 10 */
11#include <linux/types.h> 11#include <linux/types.h>
12 12
13#include <asm/hw_irq.h>
14
13#define MAX_HWCOUNTERS 8 15#define MAX_HWCOUNTERS 8
14#define MAX_EVENT_ALTERNATIVES 8 16#define MAX_EVENT_ALTERNATIVES 8
15#define MAX_LIMITED_HWCOUNTERS 2 17#define MAX_LIMITED_HWCOUNTERS 2
@@ -19,27 +21,27 @@
19 * describe the PMU on a particular POWER-family CPU. 21 * describe the PMU on a particular POWER-family CPU.
20 */ 22 */
21struct power_pmu { 23struct power_pmu {
22 int n_counter; 24 const char *name;
23 int max_alternatives; 25 int n_counter;
24 u64 add_fields; 26 int max_alternatives;
25 u64 test_adder; 27 unsigned long add_fields;
26 int (*compute_mmcr)(u64 events[], int n_ev, 28 unsigned long test_adder;
27 unsigned int hwc[], u64 mmcr[]); 29 int (*compute_mmcr)(u64 events[], int n_ev,
28 int (*get_constraint)(u64 event, u64 *mskp, u64 *valp); 30 unsigned int hwc[], unsigned long mmcr[]);
29 int (*get_alternatives)(u64 event, unsigned int flags, 31 int (*get_constraint)(u64 event, unsigned long *mskp,
30 u64 alt[]); 32 unsigned long *valp);
31 void (*disable_pmc)(unsigned int pmc, u64 mmcr[]); 33 int (*get_alternatives)(u64 event, unsigned int flags,
32 int (*limited_pmc_event)(u64 event); 34 u64 alt[]);
33 u32 flags; 35 void (*disable_pmc)(unsigned int pmc, unsigned long mmcr[]);
34 int n_generic; 36 int (*limited_pmc_event)(u64 event);
35 int *generic_events; 37 u32 flags;
36 int (*cache_events)[PERF_COUNT_HW_CACHE_MAX] 38 int n_generic;
39 int *generic_events;
40 int (*cache_events)[PERF_COUNT_HW_CACHE_MAX]
37 [PERF_COUNT_HW_CACHE_OP_MAX] 41 [PERF_COUNT_HW_CACHE_OP_MAX]
38 [PERF_COUNT_HW_CACHE_RESULT_MAX]; 42 [PERF_COUNT_HW_CACHE_RESULT_MAX];
39}; 43};
40 44
41extern struct power_pmu *ppmu;
42
43/* 45/*
44 * Values for power_pmu.flags 46 * Values for power_pmu.flags
45 */ 47 */
@@ -53,15 +55,23 @@ extern struct power_pmu *ppmu;
53#define PPMU_LIMITED_PMC_REQD 2 /* have to put this on a limited PMC */ 55#define PPMU_LIMITED_PMC_REQD 2 /* have to put this on a limited PMC */
54#define PPMU_ONLY_COUNT_RUN 4 /* only counting in run state */ 56#define PPMU_ONLY_COUNT_RUN 4 /* only counting in run state */
55 57
58extern int register_power_pmu(struct power_pmu *);
59
56struct pt_regs; 60struct pt_regs;
57extern unsigned long perf_misc_flags(struct pt_regs *regs); 61extern unsigned long perf_misc_flags(struct pt_regs *regs);
58#define perf_misc_flags(regs) perf_misc_flags(regs)
59
60extern unsigned long perf_instruction_pointer(struct pt_regs *regs); 62extern unsigned long perf_instruction_pointer(struct pt_regs *regs);
61 63
62/* 64/*
63 * The power_pmu.get_constraint function returns a 64-bit value and 65 * Only override the default definitions in include/linux/perf_counter.h
64 * a 64-bit mask that express the constraints between this event and 66 * if we have hardware PMU support.
67 */
68#ifdef CONFIG_PPC_PERF_CTRS
69#define perf_misc_flags(regs) perf_misc_flags(regs)
70#endif
71
72/*
73 * The power_pmu.get_constraint function returns a 32/64-bit value and
74 * a 32/64-bit mask that express the constraints between this event and
65 * other events. 75 * other events.
66 * 76 *
67 * The value and mask are divided up into (non-overlapping) bitfields 77 * The value and mask are divided up into (non-overlapping) bitfields
diff --git a/arch/powerpc/kernel/Makefile b/arch/powerpc/kernel/Makefile
index 6a4fb29a0618..b73396b93905 100644
--- a/arch/powerpc/kernel/Makefile
+++ b/arch/powerpc/kernel/Makefile
@@ -97,9 +97,10 @@ obj64-$(CONFIG_AUDIT) += compat_audit.o
97 97
98obj-$(CONFIG_DYNAMIC_FTRACE) += ftrace.o 98obj-$(CONFIG_DYNAMIC_FTRACE) += ftrace.o
99obj-$(CONFIG_FUNCTION_GRAPH_TRACER) += ftrace.o 99obj-$(CONFIG_FUNCTION_GRAPH_TRACER) += ftrace.o
100obj-$(CONFIG_PERF_COUNTERS) += perf_counter.o power4-pmu.o ppc970-pmu.o \ 100obj-$(CONFIG_PPC_PERF_CTRS) += perf_counter.o
101 power5-pmu.o power5+-pmu.o power6-pmu.o \ 101obj64-$(CONFIG_PPC_PERF_CTRS) += power4-pmu.o ppc970-pmu.o power5-pmu.o \
102 power7-pmu.o 102 power5+-pmu.o power6-pmu.o power7-pmu.o
103obj32-$(CONFIG_PPC_PERF_CTRS) += mpc7450-pmu.o
103 104
104obj-$(CONFIG_8XX_MINIMAL_FPEMU) += softemu8xx.o 105obj-$(CONFIG_8XX_MINIMAL_FPEMU) += softemu8xx.o
105 106
@@ -108,6 +109,7 @@ obj-y += iomap.o
108endif 109endif
109 110
110obj-$(CONFIG_PPC64) += $(obj64-y) 111obj-$(CONFIG_PPC64) += $(obj64-y)
112obj-$(CONFIG_PPC32) += $(obj32-y)
111 113
112ifneq ($(CONFIG_XMON)$(CONFIG_KEXEC),) 114ifneq ($(CONFIG_XMON)$(CONFIG_KEXEC),)
113obj-y += ppc_save_regs.o 115obj-y += ppc_save_regs.o
diff --git a/arch/powerpc/kernel/mpc7450-pmu.c b/arch/powerpc/kernel/mpc7450-pmu.c
new file mode 100644
index 000000000000..75ff47fed7bf
--- /dev/null
+++ b/arch/powerpc/kernel/mpc7450-pmu.c
@@ -0,0 +1,417 @@
1/*
2 * Performance counter support for MPC7450-family processors.
3 *
4 * Copyright 2008-2009 Paul Mackerras, IBM Corporation.
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 */
11#include <linux/string.h>
12#include <linux/perf_counter.h>
13#include <linux/string.h>
14#include <asm/reg.h>
15#include <asm/cputable.h>
16
17#define N_COUNTER 6 /* Number of hardware counters */
18#define MAX_ALT 3 /* Maximum number of event alternative codes */
19
20/*
21 * Bits in event code for MPC7450 family
22 */
23#define PM_THRMULT_MSKS 0x40000
24#define PM_THRESH_SH 12
25#define PM_THRESH_MSK 0x3f
26#define PM_PMC_SH 8
27#define PM_PMC_MSK 7
28#define PM_PMCSEL_MSK 0x7f
29
30/*
31 * Classify events according to how specific their PMC requirements are.
32 * Result is:
33 * 0: can go on any PMC
34 * 1: can go on PMCs 1-4
35 * 2: can go on PMCs 1,2,4
36 * 3: can go on PMCs 1 or 2
37 * 4: can only go on one PMC
38 * -1: event code is invalid
39 */
40#define N_CLASSES 5
41
42static int mpc7450_classify_event(u32 event)
43{
44 int pmc;
45
46 pmc = (event >> PM_PMC_SH) & PM_PMC_MSK;
47 if (pmc) {
48 if (pmc > N_COUNTER)
49 return -1;
50 return 4;
51 }
52 event &= PM_PMCSEL_MSK;
53 if (event <= 1)
54 return 0;
55 if (event <= 7)
56 return 1;
57 if (event <= 13)
58 return 2;
59 if (event <= 22)
60 return 3;
61 return -1;
62}
63
64/*
65 * Events using threshold and possible threshold scale:
66 * code scale? name
67 * 11e N PM_INSTQ_EXCEED_CYC
68 * 11f N PM_ALTV_IQ_EXCEED_CYC
69 * 128 Y PM_DTLB_SEARCH_EXCEED_CYC
70 * 12b Y PM_LD_MISS_EXCEED_L1_CYC
71 * 220 N PM_CQ_EXCEED_CYC
72 * 30c N PM_GPR_RB_EXCEED_CYC
73 * 30d ? PM_FPR_IQ_EXCEED_CYC ?
74 * 311 Y PM_ITLB_SEARCH_EXCEED
75 * 410 N PM_GPR_IQ_EXCEED_CYC
76 */
77
78/*
79 * Return use of threshold and threshold scale bits:
80 * 0 = uses neither, 1 = uses threshold, 2 = uses both
81 */
82static int mpc7450_threshold_use(u32 event)
83{
84 int pmc, sel;
85
86 pmc = (event >> PM_PMC_SH) & PM_PMC_MSK;
87 sel = event & PM_PMCSEL_MSK;
88 switch (pmc) {
89 case 1:
90 if (sel == 0x1e || sel == 0x1f)
91 return 1;
92 if (sel == 0x28 || sel == 0x2b)
93 return 2;
94 break;
95 case 2:
96 if (sel == 0x20)
97 return 1;
98 break;
99 case 3:
100 if (sel == 0xc || sel == 0xd)
101 return 1;
102 if (sel == 0x11)
103 return 2;
104 break;
105 case 4:
106 if (sel == 0x10)
107 return 1;
108 break;
109 }
110 return 0;
111}
112
113/*
114 * Layout of constraint bits:
115 * 33222222222211111111110000000000
116 * 10987654321098765432109876543210
117 * |< >< > < > < ><><><><><><>
118 * TS TV G4 G3 G2P6P5P4P3P2P1
119 *
120 * P1 - P6
121 * 0 - 11: Count of events needing PMC1 .. PMC6
122 *
123 * G2
124 * 12 - 14: Count of events needing PMC1 or PMC2
125 *
126 * G3
127 * 16 - 18: Count of events needing PMC1, PMC2 or PMC4
128 *
129 * G4
130 * 20 - 23: Count of events needing PMC1, PMC2, PMC3 or PMC4
131 *
132 * TV
133 * 24 - 29: Threshold value requested
134 *
135 * TS
136 * 30: Threshold scale value requested
137 */
138
139static u32 pmcbits[N_COUNTER][2] = {
140 { 0x00844002, 0x00111001 }, /* PMC1 mask, value: P1,G2,G3,G4 */
141 { 0x00844008, 0x00111004 }, /* PMC2: P2,G2,G3,G4 */
142 { 0x00800020, 0x00100010 }, /* PMC3: P3,G4 */
143 { 0x00840080, 0x00110040 }, /* PMC4: P4,G3,G4 */
144 { 0x00000200, 0x00000100 }, /* PMC5: P5 */
145 { 0x00000800, 0x00000400 } /* PMC6: P6 */
146};
147
148static u32 classbits[N_CLASSES - 1][2] = {
149 { 0x00000000, 0x00000000 }, /* class 0: no constraint */
150 { 0x00800000, 0x00100000 }, /* class 1: G4 */
151 { 0x00040000, 0x00010000 }, /* class 2: G3 */
152 { 0x00004000, 0x00001000 }, /* class 3: G2 */
153};
154
155static int mpc7450_get_constraint(u64 event, unsigned long *maskp,
156 unsigned long *valp)
157{
158 int pmc, class;
159 u32 mask, value;
160 int thresh, tuse;
161
162 class = mpc7450_classify_event(event);
163 if (class < 0)
164 return -1;
165 if (class == 4) {
166 pmc = ((unsigned int)event >> PM_PMC_SH) & PM_PMC_MSK;
167 mask = pmcbits[pmc - 1][0];
168 value = pmcbits[pmc - 1][1];
169 } else {
170 mask = classbits[class][0];
171 value = classbits[class][1];
172 }
173
174 tuse = mpc7450_threshold_use(event);
175 if (tuse) {
176 thresh = ((unsigned int)event >> PM_THRESH_SH) & PM_THRESH_MSK;
177 mask |= 0x3f << 24;
178 value |= thresh << 24;
179 if (tuse == 2) {
180 mask |= 0x40000000;
181 if ((unsigned int)event & PM_THRMULT_MSKS)
182 value |= 0x40000000;
183 }
184 }
185
186 *maskp = mask;
187 *valp = value;
188 return 0;
189}
190
191static const unsigned int event_alternatives[][MAX_ALT] = {
192 { 0x217, 0x317 }, /* PM_L1_DCACHE_MISS */
193 { 0x418, 0x50f, 0x60f }, /* PM_SNOOP_RETRY */
194 { 0x502, 0x602 }, /* PM_L2_HIT */
195 { 0x503, 0x603 }, /* PM_L3_HIT */
196 { 0x504, 0x604 }, /* PM_L2_ICACHE_MISS */
197 { 0x505, 0x605 }, /* PM_L3_ICACHE_MISS */
198 { 0x506, 0x606 }, /* PM_L2_DCACHE_MISS */
199 { 0x507, 0x607 }, /* PM_L3_DCACHE_MISS */
200 { 0x50a, 0x623 }, /* PM_LD_HIT_L3 */
201 { 0x50b, 0x624 }, /* PM_ST_HIT_L3 */
202 { 0x50d, 0x60d }, /* PM_L2_TOUCH_HIT */
203 { 0x50e, 0x60e }, /* PM_L3_TOUCH_HIT */
204 { 0x512, 0x612 }, /* PM_INT_LOCAL */
205 { 0x513, 0x61d }, /* PM_L2_MISS */
206 { 0x514, 0x61e }, /* PM_L3_MISS */
207};
208
209/*
210 * Scan the alternatives table for a match and return the
211 * index into the alternatives table if found, else -1.
212 */
213static int find_alternative(u32 event)
214{
215 int i, j;
216
217 for (i = 0; i < ARRAY_SIZE(event_alternatives); ++i) {
218 if (event < event_alternatives[i][0])
219 break;
220 for (j = 0; j < MAX_ALT && event_alternatives[i][j]; ++j)
221 if (event == event_alternatives[i][j])
222 return i;
223 }
224 return -1;
225}
226
227static int mpc7450_get_alternatives(u64 event, unsigned int flags, u64 alt[])
228{
229 int i, j, nalt = 1;
230 u32 ae;
231
232 alt[0] = event;
233 nalt = 1;
234 i = find_alternative((u32)event);
235 if (i >= 0) {
236 for (j = 0; j < MAX_ALT; ++j) {
237 ae = event_alternatives[i][j];
238 if (ae && ae != (u32)event)
239 alt[nalt++] = ae;
240 }
241 }
242 return nalt;
243}
244
245/*
246 * Bitmaps of which PMCs each class can use for classes 0 - 3.
247 * Bit i is set if PMC i+1 is usable.
248 */
249static const u8 classmap[N_CLASSES] = {
250 0x3f, 0x0f, 0x0b, 0x03, 0
251};
252
253/* Bit position and width of each PMCSEL field */
254static const int pmcsel_shift[N_COUNTER] = {
255 6, 0, 27, 22, 17, 11
256};
257static const u32 pmcsel_mask[N_COUNTER] = {
258 0x7f, 0x3f, 0x1f, 0x1f, 0x1f, 0x3f
259};
260
261/*
262 * Compute MMCR0/1/2 values for a set of events.
263 */
264static int mpc7450_compute_mmcr(u64 event[], int n_ev,
265 unsigned int hwc[], unsigned long mmcr[])
266{
267 u8 event_index[N_CLASSES][N_COUNTER];
268 int n_classevent[N_CLASSES];
269 int i, j, class, tuse;
270 u32 pmc_inuse = 0, pmc_avail;
271 u32 mmcr0 = 0, mmcr1 = 0, mmcr2 = 0;
272 u32 ev, pmc, thresh;
273
274 if (n_ev > N_COUNTER)
275 return -1;
276
277 /* First pass: count usage in each class */
278 for (i = 0; i < N_CLASSES; ++i)
279 n_classevent[i] = 0;
280 for (i = 0; i < n_ev; ++i) {
281 class = mpc7450_classify_event(event[i]);
282 if (class < 0)
283 return -1;
284 j = n_classevent[class]++;
285 event_index[class][j] = i;
286 }
287
288 /* Second pass: allocate PMCs from most specific event to least */
289 for (class = N_CLASSES - 1; class >= 0; --class) {
290 for (i = 0; i < n_classevent[class]; ++i) {
291 ev = event[event_index[class][i]];
292 if (class == 4) {
293 pmc = (ev >> PM_PMC_SH) & PM_PMC_MSK;
294 if (pmc_inuse & (1 << (pmc - 1)))
295 return -1;
296 } else {
297 /* Find a suitable PMC */
298 pmc_avail = classmap[class] & ~pmc_inuse;
299 if (!pmc_avail)
300 return -1;
301 pmc = ffs(pmc_avail);
302 }
303 pmc_inuse |= 1 << (pmc - 1);
304
305 tuse = mpc7450_threshold_use(ev);
306 if (tuse) {
307 thresh = (ev >> PM_THRESH_SH) & PM_THRESH_MSK;
308 mmcr0 |= thresh << 16;
309 if (tuse == 2 && (ev & PM_THRMULT_MSKS))
310 mmcr2 = 0x80000000;
311 }
312 ev &= pmcsel_mask[pmc - 1];
313 ev <<= pmcsel_shift[pmc - 1];
314 if (pmc <= 2)
315 mmcr0 |= ev;
316 else
317 mmcr1 |= ev;
318 hwc[event_index[class][i]] = pmc - 1;
319 }
320 }
321
322 if (pmc_inuse & 1)
323 mmcr0 |= MMCR0_PMC1CE;
324 if (pmc_inuse & 0x3e)
325 mmcr0 |= MMCR0_PMCnCE;
326
327 /* Return MMCRx values */
328 mmcr[0] = mmcr0;
329 mmcr[1] = mmcr1;
330 mmcr[2] = mmcr2;
331 return 0;
332}
333
334/*
335 * Disable counting by a PMC.
336 * Note that the pmc argument is 0-based here, not 1-based.
337 */
338static void mpc7450_disable_pmc(unsigned int pmc, unsigned long mmcr[])
339{
340 if (pmc <= 1)
341 mmcr[0] &= ~(pmcsel_mask[pmc] << pmcsel_shift[pmc]);
342 else
343 mmcr[1] &= ~(pmcsel_mask[pmc] << pmcsel_shift[pmc]);
344}
345
346static int mpc7450_generic_events[] = {
347 [PERF_COUNT_HW_CPU_CYCLES] = 1,
348 [PERF_COUNT_HW_INSTRUCTIONS] = 2,
349 [PERF_COUNT_HW_CACHE_MISSES] = 0x217, /* PM_L1_DCACHE_MISS */
350 [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = 0x122, /* PM_BR_CMPL */
351 [PERF_COUNT_HW_BRANCH_MISSES] = 0x41c, /* PM_BR_MPRED */
352};
353
354#define C(x) PERF_COUNT_HW_CACHE_##x
355
356/*
357 * Table of generalized cache-related events.
358 * 0 means not supported, -1 means nonsensical, other values
359 * are event codes.
360 */
361static int mpc7450_cache_events[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = {
362 [C(L1D)] = { /* RESULT_ACCESS RESULT_MISS */
363 [C(OP_READ)] = { 0, 0x225 },
364 [C(OP_WRITE)] = { 0, 0x227 },
365 [C(OP_PREFETCH)] = { 0, 0 },
366 },
367 [C(L1I)] = { /* RESULT_ACCESS RESULT_MISS */
368 [C(OP_READ)] = { 0x129, 0x115 },
369 [C(OP_WRITE)] = { -1, -1 },
370 [C(OP_PREFETCH)] = { 0x634, 0 },
371 },
372 [C(LL)] = { /* RESULT_ACCESS RESULT_MISS */
373 [C(OP_READ)] = { 0, 0 },
374 [C(OP_WRITE)] = { 0, 0 },
375 [C(OP_PREFETCH)] = { 0, 0 },
376 },
377 [C(DTLB)] = { /* RESULT_ACCESS RESULT_MISS */
378 [C(OP_READ)] = { 0, 0x312 },
379 [C(OP_WRITE)] = { -1, -1 },
380 [C(OP_PREFETCH)] = { -1, -1 },
381 },
382 [C(ITLB)] = { /* RESULT_ACCESS RESULT_MISS */
383 [C(OP_READ)] = { 0, 0x223 },
384 [C(OP_WRITE)] = { -1, -1 },
385 [C(OP_PREFETCH)] = { -1, -1 },
386 },
387 [C(BPU)] = { /* RESULT_ACCESS RESULT_MISS */
388 [C(OP_READ)] = { 0x122, 0x41c },
389 [C(OP_WRITE)] = { -1, -1 },
390 [C(OP_PREFETCH)] = { -1, -1 },
391 },
392};
393
394struct power_pmu mpc7450_pmu = {
395 .name = "MPC7450 family",
396 .n_counter = N_COUNTER,
397 .max_alternatives = MAX_ALT,
398 .add_fields = 0x00111555ul,
399 .test_adder = 0x00301000ul,
400 .compute_mmcr = mpc7450_compute_mmcr,
401 .get_constraint = mpc7450_get_constraint,
402 .get_alternatives = mpc7450_get_alternatives,
403 .disable_pmc = mpc7450_disable_pmc,
404 .n_generic = ARRAY_SIZE(mpc7450_generic_events),
405 .generic_events = mpc7450_generic_events,
406 .cache_events = &mpc7450_cache_events,
407};
408
409static int init_mpc7450_pmu(void)
410{
411 if (strcmp(cur_cpu_spec->oprofile_cpu_type, "ppc/7450"))
412 return -ENODEV;
413
414 return register_power_pmu(&mpc7450_pmu);
415}
416
417arch_initcall(init_mpc7450_pmu);
diff --git a/arch/powerpc/kernel/perf_counter.c b/arch/powerpc/kernel/perf_counter.c
index bb202388170e..809fdf94b95f 100644
--- a/arch/powerpc/kernel/perf_counter.c
+++ b/arch/powerpc/kernel/perf_counter.c
@@ -29,7 +29,7 @@ struct cpu_hw_counters {
29 struct perf_counter *counter[MAX_HWCOUNTERS]; 29 struct perf_counter *counter[MAX_HWCOUNTERS];
30 u64 events[MAX_HWCOUNTERS]; 30 u64 events[MAX_HWCOUNTERS];
31 unsigned int flags[MAX_HWCOUNTERS]; 31 unsigned int flags[MAX_HWCOUNTERS];
32 u64 mmcr[3]; 32 unsigned long mmcr[3];
33 struct perf_counter *limited_counter[MAX_LIMITED_HWCOUNTERS]; 33 struct perf_counter *limited_counter[MAX_LIMITED_HWCOUNTERS];
34 u8 limited_hwidx[MAX_LIMITED_HWCOUNTERS]; 34 u8 limited_hwidx[MAX_LIMITED_HWCOUNTERS];
35}; 35};
@@ -46,6 +46,115 @@ struct power_pmu *ppmu;
46 */ 46 */
47static unsigned int freeze_counters_kernel = MMCR0_FCS; 47static unsigned int freeze_counters_kernel = MMCR0_FCS;
48 48
49/*
50 * 32-bit doesn't have MMCRA but does have an MMCR2,
51 * and a few other names are different.
52 */
53#ifdef CONFIG_PPC32
54
55#define MMCR0_FCHV 0
56#define MMCR0_PMCjCE MMCR0_PMCnCE
57
58#define SPRN_MMCRA SPRN_MMCR2
59#define MMCRA_SAMPLE_ENABLE 0
60
61static inline unsigned long perf_ip_adjust(struct pt_regs *regs)
62{
63 return 0;
64}
65static inline void perf_set_pmu_inuse(int inuse) { }
66static inline void perf_get_data_addr(struct pt_regs *regs, u64 *addrp) { }
67static inline u32 perf_get_misc_flags(struct pt_regs *regs)
68{
69 return 0;
70}
71static inline void perf_read_regs(struct pt_regs *regs) { }
72static inline int perf_intr_is_nmi(struct pt_regs *regs)
73{
74 return 0;
75}
76
77#endif /* CONFIG_PPC32 */
78
79/*
80 * Things that are specific to 64-bit implementations.
81 */
82#ifdef CONFIG_PPC64
83
84static inline unsigned long perf_ip_adjust(struct pt_regs *regs)
85{
86 unsigned long mmcra = regs->dsisr;
87
88 if ((mmcra & MMCRA_SAMPLE_ENABLE) && !(ppmu->flags & PPMU_ALT_SIPR)) {
89 unsigned long slot = (mmcra & MMCRA_SLOT) >> MMCRA_SLOT_SHIFT;
90 if (slot > 1)
91 return 4 * (slot - 1);
92 }
93 return 0;
94}
95
96static inline void perf_set_pmu_inuse(int inuse)
97{
98 get_lppaca()->pmcregs_in_use = inuse;
99}
100
101/*
102 * The user wants a data address recorded.
103 * If we're not doing instruction sampling, give them the SDAR
104 * (sampled data address). If we are doing instruction sampling, then
105 * only give them the SDAR if it corresponds to the instruction
106 * pointed to by SIAR; this is indicated by the [POWER6_]MMCRA_SDSYNC
107 * bit in MMCRA.
108 */
109static inline void perf_get_data_addr(struct pt_regs *regs, u64 *addrp)
110{
111 unsigned long mmcra = regs->dsisr;
112 unsigned long sdsync = (ppmu->flags & PPMU_ALT_SIPR) ?
113 POWER6_MMCRA_SDSYNC : MMCRA_SDSYNC;
114
115 if (!(mmcra & MMCRA_SAMPLE_ENABLE) || (mmcra & sdsync))
116 *addrp = mfspr(SPRN_SDAR);
117}
118
119static inline u32 perf_get_misc_flags(struct pt_regs *regs)
120{
121 unsigned long mmcra = regs->dsisr;
122
123 if (TRAP(regs) != 0xf00)
124 return 0; /* not a PMU interrupt */
125
126 if (ppmu->flags & PPMU_ALT_SIPR) {
127 if (mmcra & POWER6_MMCRA_SIHV)
128 return PERF_EVENT_MISC_HYPERVISOR;
129 return (mmcra & POWER6_MMCRA_SIPR) ?
130 PERF_EVENT_MISC_USER : PERF_EVENT_MISC_KERNEL;
131 }
132 if (mmcra & MMCRA_SIHV)
133 return PERF_EVENT_MISC_HYPERVISOR;
134 return (mmcra & MMCRA_SIPR) ? PERF_EVENT_MISC_USER :
135 PERF_EVENT_MISC_KERNEL;
136}
137
138/*
139 * Overload regs->dsisr to store MMCRA so we only need to read it once
140 * on each interrupt.
141 */
142static inline void perf_read_regs(struct pt_regs *regs)
143{
144 regs->dsisr = mfspr(SPRN_MMCRA);
145}
146
147/*
148 * If interrupts were soft-disabled when a PMU interrupt occurs, treat
149 * it as an NMI.
150 */
151static inline int perf_intr_is_nmi(struct pt_regs *regs)
152{
153 return !regs->softe;
154}
155
156#endif /* CONFIG_PPC64 */
157
49static void perf_counter_interrupt(struct pt_regs *regs); 158static void perf_counter_interrupt(struct pt_regs *regs);
50 159
51void perf_counter_print_debug(void) 160void perf_counter_print_debug(void)
@@ -78,12 +187,14 @@ static unsigned long read_pmc(int idx)
78 case 6: 187 case 6:
79 val = mfspr(SPRN_PMC6); 188 val = mfspr(SPRN_PMC6);
80 break; 189 break;
190#ifdef CONFIG_PPC64
81 case 7: 191 case 7:
82 val = mfspr(SPRN_PMC7); 192 val = mfspr(SPRN_PMC7);
83 break; 193 break;
84 case 8: 194 case 8:
85 val = mfspr(SPRN_PMC8); 195 val = mfspr(SPRN_PMC8);
86 break; 196 break;
197#endif /* CONFIG_PPC64 */
87 default: 198 default:
88 printk(KERN_ERR "oops trying to read PMC%d\n", idx); 199 printk(KERN_ERR "oops trying to read PMC%d\n", idx);
89 val = 0; 200 val = 0;
@@ -115,12 +226,14 @@ static void write_pmc(int idx, unsigned long val)
115 case 6: 226 case 6:
116 mtspr(SPRN_PMC6, val); 227 mtspr(SPRN_PMC6, val);
117 break; 228 break;
229#ifdef CONFIG_PPC64
118 case 7: 230 case 7:
119 mtspr(SPRN_PMC7, val); 231 mtspr(SPRN_PMC7, val);
120 break; 232 break;
121 case 8: 233 case 8:
122 mtspr(SPRN_PMC8, val); 234 mtspr(SPRN_PMC8, val);
123 break; 235 break;
236#endif /* CONFIG_PPC64 */
124 default: 237 default:
125 printk(KERN_ERR "oops trying to write PMC%d\n", idx); 238 printk(KERN_ERR "oops trying to write PMC%d\n", idx);
126 } 239 }
@@ -135,15 +248,15 @@ static void write_pmc(int idx, unsigned long val)
135static int power_check_constraints(u64 event[], unsigned int cflags[], 248static int power_check_constraints(u64 event[], unsigned int cflags[],
136 int n_ev) 249 int n_ev)
137{ 250{
138 u64 mask, value, nv; 251 unsigned long mask, value, nv;
139 u64 alternatives[MAX_HWCOUNTERS][MAX_EVENT_ALTERNATIVES]; 252 u64 alternatives[MAX_HWCOUNTERS][MAX_EVENT_ALTERNATIVES];
140 u64 amasks[MAX_HWCOUNTERS][MAX_EVENT_ALTERNATIVES]; 253 unsigned long amasks[MAX_HWCOUNTERS][MAX_EVENT_ALTERNATIVES];
141 u64 avalues[MAX_HWCOUNTERS][MAX_EVENT_ALTERNATIVES]; 254 unsigned long avalues[MAX_HWCOUNTERS][MAX_EVENT_ALTERNATIVES];
142 u64 smasks[MAX_HWCOUNTERS], svalues[MAX_HWCOUNTERS]; 255 unsigned long smasks[MAX_HWCOUNTERS], svalues[MAX_HWCOUNTERS];
143 int n_alt[MAX_HWCOUNTERS], choice[MAX_HWCOUNTERS]; 256 int n_alt[MAX_HWCOUNTERS], choice[MAX_HWCOUNTERS];
144 int i, j; 257 int i, j;
145 u64 addf = ppmu->add_fields; 258 unsigned long addf = ppmu->add_fields;
146 u64 tadd = ppmu->test_adder; 259 unsigned long tadd = ppmu->test_adder;
147 260
148 if (n_ev > ppmu->n_counter) 261 if (n_ev > ppmu->n_counter)
149 return -1; 262 return -1;
@@ -283,7 +396,7 @@ static int check_excludes(struct perf_counter **ctrs, unsigned int cflags[],
283 396
284static void power_pmu_read(struct perf_counter *counter) 397static void power_pmu_read(struct perf_counter *counter)
285{ 398{
286 long val, delta, prev; 399 s64 val, delta, prev;
287 400
288 if (!counter->hw.idx) 401 if (!counter->hw.idx)
289 return; 402 return;
@@ -403,14 +516,12 @@ static void write_mmcr0(struct cpu_hw_counters *cpuhw, unsigned long mmcr0)
403void hw_perf_disable(void) 516void hw_perf_disable(void)
404{ 517{
405 struct cpu_hw_counters *cpuhw; 518 struct cpu_hw_counters *cpuhw;
406 unsigned long ret;
407 unsigned long flags; 519 unsigned long flags;
408 520
409 local_irq_save(flags); 521 local_irq_save(flags);
410 cpuhw = &__get_cpu_var(cpu_hw_counters); 522 cpuhw = &__get_cpu_var(cpu_hw_counters);
411 523
412 ret = cpuhw->disabled; 524 if (!cpuhw->disabled) {
413 if (!ret) {
414 cpuhw->disabled = 1; 525 cpuhw->disabled = 1;
415 cpuhw->n_added = 0; 526 cpuhw->n_added = 0;
416 527
@@ -479,7 +590,7 @@ void hw_perf_enable(void)
479 mtspr(SPRN_MMCRA, cpuhw->mmcr[2] & ~MMCRA_SAMPLE_ENABLE); 590 mtspr(SPRN_MMCRA, cpuhw->mmcr[2] & ~MMCRA_SAMPLE_ENABLE);
480 mtspr(SPRN_MMCR1, cpuhw->mmcr[1]); 591 mtspr(SPRN_MMCR1, cpuhw->mmcr[1]);
481 if (cpuhw->n_counters == 0) 592 if (cpuhw->n_counters == 0)
482 get_lppaca()->pmcregs_in_use = 0; 593 perf_set_pmu_inuse(0);
483 goto out_enable; 594 goto out_enable;
484 } 595 }
485 596
@@ -512,7 +623,7 @@ void hw_perf_enable(void)
512 * bit set and set the hardware counters to their initial values. 623 * bit set and set the hardware counters to their initial values.
513 * Then unfreeze the counters. 624 * Then unfreeze the counters.
514 */ 625 */
515 get_lppaca()->pmcregs_in_use = 1; 626 perf_set_pmu_inuse(1);
516 mtspr(SPRN_MMCRA, cpuhw->mmcr[2] & ~MMCRA_SAMPLE_ENABLE); 627 mtspr(SPRN_MMCRA, cpuhw->mmcr[2] & ~MMCRA_SAMPLE_ENABLE);
517 mtspr(SPRN_MMCR1, cpuhw->mmcr[1]); 628 mtspr(SPRN_MMCR1, cpuhw->mmcr[1]);
518 mtspr(SPRN_MMCR0, (cpuhw->mmcr[0] & ~(MMCR0_PMC1CE | MMCR0_PMCjCE)) 629 mtspr(SPRN_MMCR0, (cpuhw->mmcr[0] & ~(MMCR0_PMC1CE | MMCR0_PMCjCE))
@@ -913,6 +1024,8 @@ const struct pmu *hw_perf_counter_init(struct perf_counter *counter)
913 case PERF_TYPE_RAW: 1024 case PERF_TYPE_RAW:
914 ev = counter->attr.config; 1025 ev = counter->attr.config;
915 break; 1026 break;
1027 default:
1028 return ERR_PTR(-EINVAL);
916 } 1029 }
917 counter->hw.config_base = ev; 1030 counter->hw.config_base = ev;
918 counter->hw.idx = 0; 1031 counter->hw.idx = 0;
@@ -1007,13 +1120,12 @@ const struct pmu *hw_perf_counter_init(struct perf_counter *counter)
1007 * things if requested. Note that interrupts are hard-disabled 1120 * things if requested. Note that interrupts are hard-disabled
1008 * here so there is no possibility of being interrupted. 1121 * here so there is no possibility of being interrupted.
1009 */ 1122 */
1010static void record_and_restart(struct perf_counter *counter, long val, 1123static void record_and_restart(struct perf_counter *counter, unsigned long val,
1011 struct pt_regs *regs, int nmi) 1124 struct pt_regs *regs, int nmi)
1012{ 1125{
1013 u64 period = counter->hw.sample_period; 1126 u64 period = counter->hw.sample_period;
1014 s64 prev, delta, left; 1127 s64 prev, delta, left;
1015 int record = 0; 1128 int record = 0;
1016 u64 addr, mmcra, sdsync;
1017 1129
1018 /* we don't have to worry about interrupts here */ 1130 /* we don't have to worry about interrupts here */
1019 prev = atomic64_read(&counter->hw.prev_count); 1131 prev = atomic64_read(&counter->hw.prev_count);
@@ -1033,8 +1145,8 @@ static void record_and_restart(struct perf_counter *counter, long val,
1033 left = period; 1145 left = period;
1034 record = 1; 1146 record = 1;
1035 } 1147 }
1036 if (left < 0x80000000L) 1148 if (left < 0x80000000LL)
1037 val = 0x80000000L - left; 1149 val = 0x80000000LL - left;
1038 } 1150 }
1039 1151
1040 /* 1152 /*
@@ -1047,22 +1159,9 @@ static void record_and_restart(struct perf_counter *counter, long val,
1047 .period = counter->hw.last_period, 1159 .period = counter->hw.last_period,
1048 }; 1160 };
1049 1161
1050 if (counter->attr.sample_type & PERF_SAMPLE_ADDR) { 1162 if (counter->attr.sample_type & PERF_SAMPLE_ADDR)
1051 /* 1163 perf_get_data_addr(regs, &data.addr);
1052 * The user wants a data address recorded. 1164
1053 * If we're not doing instruction sampling,
1054 * give them the SDAR (sampled data address).
1055 * If we are doing instruction sampling, then only
1056 * give them the SDAR if it corresponds to the
1057 * instruction pointed to by SIAR; this is indicated
1058 * by the [POWER6_]MMCRA_SDSYNC bit in MMCRA.
1059 */
1060 mmcra = regs->dsisr;
1061 sdsync = (ppmu->flags & PPMU_ALT_SIPR) ?
1062 POWER6_MMCRA_SDSYNC : MMCRA_SDSYNC;
1063 if (!(mmcra & MMCRA_SAMPLE_ENABLE) || (mmcra & sdsync))
1064 data.addr = mfspr(SPRN_SDAR);
1065 }
1066 if (perf_counter_overflow(counter, nmi, &data)) { 1165 if (perf_counter_overflow(counter, nmi, &data)) {
1067 /* 1166 /*
1068 * Interrupts are coming too fast - throttle them 1167 * Interrupts are coming too fast - throttle them
@@ -1088,25 +1187,12 @@ static void record_and_restart(struct perf_counter *counter, long val,
1088 */ 1187 */
1089unsigned long perf_misc_flags(struct pt_regs *regs) 1188unsigned long perf_misc_flags(struct pt_regs *regs)
1090{ 1189{
1091 unsigned long mmcra; 1190 u32 flags = perf_get_misc_flags(regs);
1092
1093 if (TRAP(regs) != 0xf00) {
1094 /* not a PMU interrupt */
1095 return user_mode(regs) ? PERF_EVENT_MISC_USER :
1096 PERF_EVENT_MISC_KERNEL;
1097 }
1098 1191
1099 mmcra = regs->dsisr; 1192 if (flags)
1100 if (ppmu->flags & PPMU_ALT_SIPR) { 1193 return flags;
1101 if (mmcra & POWER6_MMCRA_SIHV) 1194 return user_mode(regs) ? PERF_EVENT_MISC_USER :
1102 return PERF_EVENT_MISC_HYPERVISOR; 1195 PERF_EVENT_MISC_KERNEL;
1103 return (mmcra & POWER6_MMCRA_SIPR) ? PERF_EVENT_MISC_USER :
1104 PERF_EVENT_MISC_KERNEL;
1105 }
1106 if (mmcra & MMCRA_SIHV)
1107 return PERF_EVENT_MISC_HYPERVISOR;
1108 return (mmcra & MMCRA_SIPR) ? PERF_EVENT_MISC_USER :
1109 PERF_EVENT_MISC_KERNEL;
1110} 1196}
1111 1197
1112/* 1198/*
@@ -1115,20 +1201,12 @@ unsigned long perf_misc_flags(struct pt_regs *regs)
1115 */ 1201 */
1116unsigned long perf_instruction_pointer(struct pt_regs *regs) 1202unsigned long perf_instruction_pointer(struct pt_regs *regs)
1117{ 1203{
1118 unsigned long mmcra;
1119 unsigned long ip; 1204 unsigned long ip;
1120 unsigned long slot;
1121 1205
1122 if (TRAP(regs) != 0xf00) 1206 if (TRAP(regs) != 0xf00)
1123 return regs->nip; /* not a PMU interrupt */ 1207 return regs->nip; /* not a PMU interrupt */
1124 1208
1125 ip = mfspr(SPRN_SIAR); 1209 ip = mfspr(SPRN_SIAR) + perf_ip_adjust(regs);
1126 mmcra = regs->dsisr;
1127 if ((mmcra & MMCRA_SAMPLE_ENABLE) && !(ppmu->flags & PPMU_ALT_SIPR)) {
1128 slot = (mmcra & MMCRA_SLOT) >> MMCRA_SLOT_SHIFT;
1129 if (slot > 1)
1130 ip += 4 * (slot - 1);
1131 }
1132 return ip; 1210 return ip;
1133} 1211}
1134 1212
@@ -1140,7 +1218,7 @@ static void perf_counter_interrupt(struct pt_regs *regs)
1140 int i; 1218 int i;
1141 struct cpu_hw_counters *cpuhw = &__get_cpu_var(cpu_hw_counters); 1219 struct cpu_hw_counters *cpuhw = &__get_cpu_var(cpu_hw_counters);
1142 struct perf_counter *counter; 1220 struct perf_counter *counter;
1143 long val; 1221 unsigned long val;
1144 int found = 0; 1222 int found = 0;
1145 int nmi; 1223 int nmi;
1146 1224
@@ -1148,16 +1226,9 @@ static void perf_counter_interrupt(struct pt_regs *regs)
1148 freeze_limited_counters(cpuhw, mfspr(SPRN_PMC5), 1226 freeze_limited_counters(cpuhw, mfspr(SPRN_PMC5),
1149 mfspr(SPRN_PMC6)); 1227 mfspr(SPRN_PMC6));
1150 1228
1151 /* 1229 perf_read_regs(regs);
1152 * Overload regs->dsisr to store MMCRA so we only need to read it once.
1153 */
1154 regs->dsisr = mfspr(SPRN_MMCRA);
1155 1230
1156 /* 1231 nmi = perf_intr_is_nmi(regs);
1157 * If interrupts were soft-disabled when this PMU interrupt
1158 * occurred, treat it as an NMI.
1159 */
1160 nmi = !regs->softe;
1161 if (nmi) 1232 if (nmi)
1162 nmi_enter(); 1233 nmi_enter();
1163 else 1234 else
@@ -1214,50 +1285,22 @@ void hw_perf_counter_setup(int cpu)
1214 cpuhw->mmcr[0] = MMCR0_FC; 1285 cpuhw->mmcr[0] = MMCR0_FC;
1215} 1286}
1216 1287
1217extern struct power_pmu power4_pmu; 1288int register_power_pmu(struct power_pmu *pmu)
1218extern struct power_pmu ppc970_pmu;
1219extern struct power_pmu power5_pmu;
1220extern struct power_pmu power5p_pmu;
1221extern struct power_pmu power6_pmu;
1222extern struct power_pmu power7_pmu;
1223
1224static int init_perf_counters(void)
1225{ 1289{
1226 unsigned long pvr; 1290 if (ppmu)
1227 1291 return -EBUSY; /* something's already registered */
1228 /* XXX should get this from cputable */ 1292
1229 pvr = mfspr(SPRN_PVR); 1293 ppmu = pmu;
1230 switch (PVR_VER(pvr)) { 1294 pr_info("%s performance monitor hardware support registered\n",
1231 case PV_POWER4: 1295 pmu->name);
1232 case PV_POWER4p:
1233 ppmu = &power4_pmu;
1234 break;
1235 case PV_970:
1236 case PV_970FX:
1237 case PV_970MP:
1238 ppmu = &ppc970_pmu;
1239 break;
1240 case PV_POWER5:
1241 ppmu = &power5_pmu;
1242 break;
1243 case PV_POWER5p:
1244 ppmu = &power5p_pmu;
1245 break;
1246 case 0x3e:
1247 ppmu = &power6_pmu;
1248 break;
1249 case 0x3f:
1250 ppmu = &power7_pmu;
1251 break;
1252 }
1253 1296
1297#ifdef MSR_HV
1254 /* 1298 /*
1255 * Use FCHV to ignore kernel events if MSR.HV is set. 1299 * Use FCHV to ignore kernel events if MSR.HV is set.
1256 */ 1300 */
1257 if (mfmsr() & MSR_HV) 1301 if (mfmsr() & MSR_HV)
1258 freeze_counters_kernel = MMCR0_FCHV; 1302 freeze_counters_kernel = MMCR0_FCHV;
1303#endif /* CONFIG_PPC64 */
1259 1304
1260 return 0; 1305 return 0;
1261} 1306}
1262
1263arch_initcall(init_perf_counters);
diff --git a/arch/powerpc/kernel/power4-pmu.c b/arch/powerpc/kernel/power4-pmu.c
index 07bd308a5fa7..db90b0c5c27b 100644
--- a/arch/powerpc/kernel/power4-pmu.c
+++ b/arch/powerpc/kernel/power4-pmu.c
@@ -10,7 +10,9 @@
10 */ 10 */
11#include <linux/kernel.h> 11#include <linux/kernel.h>
12#include <linux/perf_counter.h> 12#include <linux/perf_counter.h>
13#include <linux/string.h>
13#include <asm/reg.h> 14#include <asm/reg.h>
15#include <asm/cputable.h>
14 16
15/* 17/*
16 * Bits in event code for POWER4 18 * Bits in event code for POWER4
@@ -179,22 +181,22 @@ static short mmcr1_adder_bits[8] = {
179 */ 181 */
180 182
181static struct unitinfo { 183static struct unitinfo {
182 u64 value, mask; 184 unsigned long value, mask;
183 int unit; 185 int unit;
184 int lowerbit; 186 int lowerbit;
185} p4_unitinfo[16] = { 187} p4_unitinfo[16] = {
186 [PM_FPU] = { 0x44000000000000ull, 0x88000000000000ull, PM_FPU, 0 }, 188 [PM_FPU] = { 0x44000000000000ul, 0x88000000000000ul, PM_FPU, 0 },
187 [PM_ISU1] = { 0x20080000000000ull, 0x88000000000000ull, PM_ISU1, 0 }, 189 [PM_ISU1] = { 0x20080000000000ul, 0x88000000000000ul, PM_ISU1, 0 },
188 [PM_ISU1_ALT] = 190 [PM_ISU1_ALT] =
189 { 0x20080000000000ull, 0x88000000000000ull, PM_ISU1, 0 }, 191 { 0x20080000000000ul, 0x88000000000000ul, PM_ISU1, 0 },
190 [PM_IFU] = { 0x02200000000000ull, 0x08820000000000ull, PM_IFU, 41 }, 192 [PM_IFU] = { 0x02200000000000ul, 0x08820000000000ul, PM_IFU, 41 },
191 [PM_IFU_ALT] = 193 [PM_IFU_ALT] =
192 { 0x02200000000000ull, 0x08820000000000ull, PM_IFU, 41 }, 194 { 0x02200000000000ul, 0x08820000000000ul, PM_IFU, 41 },
193 [PM_IDU0] = { 0x10100000000000ull, 0x80840000000000ull, PM_IDU0, 1 }, 195 [PM_IDU0] = { 0x10100000000000ul, 0x80840000000000ul, PM_IDU0, 1 },
194 [PM_ISU2] = { 0x10140000000000ull, 0x80840000000000ull, PM_ISU2, 0 }, 196 [PM_ISU2] = { 0x10140000000000ul, 0x80840000000000ul, PM_ISU2, 0 },
195 [PM_LSU0] = { 0x01400000000000ull, 0x08800000000000ull, PM_LSU0, 0 }, 197 [PM_LSU0] = { 0x01400000000000ul, 0x08800000000000ul, PM_LSU0, 0 },
196 [PM_LSU1] = { 0x00000000000000ull, 0x00010000000000ull, PM_LSU1, 40 }, 198 [PM_LSU1] = { 0x00000000000000ul, 0x00010000000000ul, PM_LSU1, 40 },
197 [PM_GPS] = { 0x00000000000000ull, 0x00000000000000ull, PM_GPS, 0 } 199 [PM_GPS] = { 0x00000000000000ul, 0x00000000000000ul, PM_GPS, 0 }
198}; 200};
199 201
200static unsigned char direct_marked_event[8] = { 202static unsigned char direct_marked_event[8] = {
@@ -249,10 +251,11 @@ static int p4_marked_instr_event(u64 event)
249 return (mask >> (byte * 8 + bit)) & 1; 251 return (mask >> (byte * 8 + bit)) & 1;
250} 252}
251 253
252static int p4_get_constraint(u64 event, u64 *maskp, u64 *valp) 254static int p4_get_constraint(u64 event, unsigned long *maskp,
255 unsigned long *valp)
253{ 256{
254 int pmc, byte, unit, lower, sh; 257 int pmc, byte, unit, lower, sh;
255 u64 mask = 0, value = 0; 258 unsigned long mask = 0, value = 0;
256 int grp = -1; 259 int grp = -1;
257 260
258 pmc = (event >> PM_PMC_SH) & PM_PMC_MSK; 261 pmc = (event >> PM_PMC_SH) & PM_PMC_MSK;
@@ -282,14 +285,14 @@ static int p4_get_constraint(u64 event, u64 *maskp, u64 *valp)
282 value |= p4_unitinfo[unit].value; 285 value |= p4_unitinfo[unit].value;
283 sh = p4_unitinfo[unit].lowerbit; 286 sh = p4_unitinfo[unit].lowerbit;
284 if (sh > 1) 287 if (sh > 1)
285 value |= (u64)lower << sh; 288 value |= (unsigned long)lower << sh;
286 else if (lower != sh) 289 else if (lower != sh)
287 return -1; 290 return -1;
288 unit = p4_unitinfo[unit].unit; 291 unit = p4_unitinfo[unit].unit;
289 292
290 /* Set byte lane select field */ 293 /* Set byte lane select field */
291 mask |= 0xfULL << (28 - 4 * byte); 294 mask |= 0xfULL << (28 - 4 * byte);
292 value |= (u64)unit << (28 - 4 * byte); 295 value |= (unsigned long)unit << (28 - 4 * byte);
293 } 296 }
294 if (grp == 0) { 297 if (grp == 0) {
295 /* increment PMC1/2/5/6 field */ 298 /* increment PMC1/2/5/6 field */
@@ -353,9 +356,9 @@ static int p4_get_alternatives(u64 event, unsigned int flags, u64 alt[])
353} 356}
354 357
355static int p4_compute_mmcr(u64 event[], int n_ev, 358static int p4_compute_mmcr(u64 event[], int n_ev,
356 unsigned int hwc[], u64 mmcr[]) 359 unsigned int hwc[], unsigned long mmcr[])
357{ 360{
358 u64 mmcr0 = 0, mmcr1 = 0, mmcra = 0; 361 unsigned long mmcr0 = 0, mmcr1 = 0, mmcra = 0;
359 unsigned int pmc, unit, byte, psel, lower; 362 unsigned int pmc, unit, byte, psel, lower;
360 unsigned int ttm, grp; 363 unsigned int ttm, grp;
361 unsigned int pmc_inuse = 0; 364 unsigned int pmc_inuse = 0;
@@ -429,9 +432,11 @@ static int p4_compute_mmcr(u64 event[], int n_ev,
429 return -1; 432 return -1;
430 433
431 /* Set TTMxSEL fields. Note, units 1-3 => TTM0SEL codes 0-2 */ 434 /* Set TTMxSEL fields. Note, units 1-3 => TTM0SEL codes 0-2 */
432 mmcr1 |= (u64)(unituse[3] * 2 + unituse[2]) << MMCR1_TTM0SEL_SH; 435 mmcr1 |= (unsigned long)(unituse[3] * 2 + unituse[2])
433 mmcr1 |= (u64)(unituse[7] * 3 + unituse[6] * 2) << MMCR1_TTM1SEL_SH; 436 << MMCR1_TTM0SEL_SH;
434 mmcr1 |= (u64)unituse[9] << MMCR1_TTM2SEL_SH; 437 mmcr1 |= (unsigned long)(unituse[7] * 3 + unituse[6] * 2)
438 << MMCR1_TTM1SEL_SH;
439 mmcr1 |= (unsigned long)unituse[9] << MMCR1_TTM2SEL_SH;
435 440
436 /* Set TTCxSEL fields. */ 441 /* Set TTCxSEL fields. */
437 if (unitlower & 0xe) 442 if (unitlower & 0xe)
@@ -456,7 +461,8 @@ static int p4_compute_mmcr(u64 event[], int n_ev,
456 ttm = unit - 1; /* 2->1, 3->2 */ 461 ttm = unit - 1; /* 2->1, 3->2 */
457 else 462 else
458 ttm = unit >> 2; 463 ttm = unit >> 2;
459 mmcr1 |= (u64)ttm << (MMCR1_TD_CP_DBG0SEL_SH - 2*byte); 464 mmcr1 |= (unsigned long)ttm
465 << (MMCR1_TD_CP_DBG0SEL_SH - 2 * byte);
460 } 466 }
461 } 467 }
462 468
@@ -519,7 +525,7 @@ static int p4_compute_mmcr(u64 event[], int n_ev,
519 return 0; 525 return 0;
520} 526}
521 527
522static void p4_disable_pmc(unsigned int pmc, u64 mmcr[]) 528static void p4_disable_pmc(unsigned int pmc, unsigned long mmcr[])
523{ 529{
524 /* 530 /*
525 * Setting the PMCxSEL field to 0 disables PMC x. 531 * Setting the PMCxSEL field to 0 disables PMC x.
@@ -583,16 +589,27 @@ static int power4_cache_events[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = {
583 }, 589 },
584}; 590};
585 591
586struct power_pmu power4_pmu = { 592static struct power_pmu power4_pmu = {
587 .n_counter = 8, 593 .name = "POWER4/4+",
588 .max_alternatives = 5, 594 .n_counter = 8,
589 .add_fields = 0x0000001100005555ull, 595 .max_alternatives = 5,
590 .test_adder = 0x0011083300000000ull, 596 .add_fields = 0x0000001100005555ul,
591 .compute_mmcr = p4_compute_mmcr, 597 .test_adder = 0x0011083300000000ul,
592 .get_constraint = p4_get_constraint, 598 .compute_mmcr = p4_compute_mmcr,
593 .get_alternatives = p4_get_alternatives, 599 .get_constraint = p4_get_constraint,
594 .disable_pmc = p4_disable_pmc, 600 .get_alternatives = p4_get_alternatives,
595 .n_generic = ARRAY_SIZE(p4_generic_events), 601 .disable_pmc = p4_disable_pmc,
596 .generic_events = p4_generic_events, 602 .n_generic = ARRAY_SIZE(p4_generic_events),
597 .cache_events = &power4_cache_events, 603 .generic_events = p4_generic_events,
604 .cache_events = &power4_cache_events,
598}; 605};
606
607static int init_power4_pmu(void)
608{
609 if (strcmp(cur_cpu_spec->oprofile_cpu_type, "ppc64/power4"))
610 return -ENODEV;
611
612 return register_power_pmu(&power4_pmu);
613}
614
615arch_initcall(init_power4_pmu);
diff --git a/arch/powerpc/kernel/power5+-pmu.c b/arch/powerpc/kernel/power5+-pmu.c
index 41e5d2d958d4..f4adca8e98a4 100644
--- a/arch/powerpc/kernel/power5+-pmu.c
+++ b/arch/powerpc/kernel/power5+-pmu.c
@@ -10,7 +10,9 @@
10 */ 10 */
11#include <linux/kernel.h> 11#include <linux/kernel.h>
12#include <linux/perf_counter.h> 12#include <linux/perf_counter.h>
13#include <linux/string.h>
13#include <asm/reg.h> 14#include <asm/reg.h>
15#include <asm/cputable.h>
14 16
15/* 17/*
16 * Bits in event code for POWER5+ (POWER5 GS) and POWER5++ (POWER5 GS DD3) 18 * Bits in event code for POWER5+ (POWER5 GS) and POWER5++ (POWER5 GS DD3)
@@ -126,20 +128,21 @@ static const int grsel_shift[8] = {
126}; 128};
127 129
128/* Masks and values for using events from the various units */ 130/* Masks and values for using events from the various units */
129static u64 unit_cons[PM_LASTUNIT+1][2] = { 131static unsigned long unit_cons[PM_LASTUNIT+1][2] = {
130 [PM_FPU] = { 0x3200000000ull, 0x0100000000ull }, 132 [PM_FPU] = { 0x3200000000ul, 0x0100000000ul },
131 [PM_ISU0] = { 0x0200000000ull, 0x0080000000ull }, 133 [PM_ISU0] = { 0x0200000000ul, 0x0080000000ul },
132 [PM_ISU1] = { 0x3200000000ull, 0x3100000000ull }, 134 [PM_ISU1] = { 0x3200000000ul, 0x3100000000ul },
133 [PM_IFU] = { 0x3200000000ull, 0x2100000000ull }, 135 [PM_IFU] = { 0x3200000000ul, 0x2100000000ul },
134 [PM_IDU] = { 0x0e00000000ull, 0x0040000000ull }, 136 [PM_IDU] = { 0x0e00000000ul, 0x0040000000ul },
135 [PM_GRS] = { 0x0e00000000ull, 0x0c40000000ull }, 137 [PM_GRS] = { 0x0e00000000ul, 0x0c40000000ul },
136}; 138};
137 139
138static int power5p_get_constraint(u64 event, u64 *maskp, u64 *valp) 140static int power5p_get_constraint(u64 event, unsigned long *maskp,
141 unsigned long *valp)
139{ 142{
140 int pmc, byte, unit, sh; 143 int pmc, byte, unit, sh;
141 int bit, fmask; 144 int bit, fmask;
142 u64 mask = 0, value = 0; 145 unsigned long mask = 0, value = 0;
143 146
144 pmc = (event >> PM_PMC_SH) & PM_PMC_MSK; 147 pmc = (event >> PM_PMC_SH) & PM_PMC_MSK;
145 if (pmc) { 148 if (pmc) {
@@ -171,17 +174,18 @@ static int power5p_get_constraint(u64 event, u64 *maskp, u64 *valp)
171 bit = event & 7; 174 bit = event & 7;
172 fmask = (bit == 6)? 7: 3; 175 fmask = (bit == 6)? 7: 3;
173 sh = grsel_shift[bit]; 176 sh = grsel_shift[bit];
174 mask |= (u64)fmask << sh; 177 mask |= (unsigned long)fmask << sh;
175 value |= (u64)((event >> PM_GRS_SH) & fmask) << sh; 178 value |= (unsigned long)((event >> PM_GRS_SH) & fmask)
179 << sh;
176 } 180 }
177 /* Set byte lane select field */ 181 /* Set byte lane select field */
178 mask |= 0xfULL << (24 - 4 * byte); 182 mask |= 0xfUL << (24 - 4 * byte);
179 value |= (u64)unit << (24 - 4 * byte); 183 value |= (unsigned long)unit << (24 - 4 * byte);
180 } 184 }
181 if (pmc < 5) { 185 if (pmc < 5) {
182 /* need a counter from PMC1-4 set */ 186 /* need a counter from PMC1-4 set */
183 mask |= 0x8000000000000ull; 187 mask |= 0x8000000000000ul;
184 value |= 0x1000000000000ull; 188 value |= 0x1000000000000ul;
185 } 189 }
186 *maskp = mask; 190 *maskp = mask;
187 *valp = value; 191 *valp = value;
@@ -452,10 +456,10 @@ static int power5p_marked_instr_event(u64 event)
452} 456}
453 457
454static int power5p_compute_mmcr(u64 event[], int n_ev, 458static int power5p_compute_mmcr(u64 event[], int n_ev,
455 unsigned int hwc[], u64 mmcr[]) 459 unsigned int hwc[], unsigned long mmcr[])
456{ 460{
457 u64 mmcr1 = 0; 461 unsigned long mmcr1 = 0;
458 u64 mmcra = 0; 462 unsigned long mmcra = 0;
459 unsigned int pmc, unit, byte, psel; 463 unsigned int pmc, unit, byte, psel;
460 unsigned int ttm; 464 unsigned int ttm;
461 int i, isbus, bit, grsel; 465 int i, isbus, bit, grsel;
@@ -517,7 +521,7 @@ static int power5p_compute_mmcr(u64 event[], int n_ev,
517 continue; 521 continue;
518 if (ttmuse++) 522 if (ttmuse++)
519 return -1; 523 return -1;
520 mmcr1 |= (u64)i << MMCR1_TTM0SEL_SH; 524 mmcr1 |= (unsigned long)i << MMCR1_TTM0SEL_SH;
521 } 525 }
522 ttmuse = 0; 526 ttmuse = 0;
523 for (; i <= PM_GRS; ++i) { 527 for (; i <= PM_GRS; ++i) {
@@ -525,7 +529,7 @@ static int power5p_compute_mmcr(u64 event[], int n_ev,
525 continue; 529 continue;
526 if (ttmuse++) 530 if (ttmuse++)
527 return -1; 531 return -1;
528 mmcr1 |= (u64)(i & 3) << MMCR1_TTM1SEL_SH; 532 mmcr1 |= (unsigned long)(i & 3) << MMCR1_TTM1SEL_SH;
529 } 533 }
530 if (ttmuse > 1) 534 if (ttmuse > 1)
531 return -1; 535 return -1;
@@ -540,10 +544,11 @@ static int power5p_compute_mmcr(u64 event[], int n_ev,
540 unit = PM_ISU0_ALT; 544 unit = PM_ISU0_ALT;
541 } else if (unit == PM_LSU1 + 1) { 545 } else if (unit == PM_LSU1 + 1) {
542 /* select lower word of LSU1 for this byte */ 546 /* select lower word of LSU1 for this byte */
543 mmcr1 |= 1ull << (MMCR1_TTM3SEL_SH + 3 - byte); 547 mmcr1 |= 1ul << (MMCR1_TTM3SEL_SH + 3 - byte);
544 } 548 }
545 ttm = unit >> 2; 549 ttm = unit >> 2;
546 mmcr1 |= (u64)ttm << (MMCR1_TD_CP_DBG0SEL_SH - 2 * byte); 550 mmcr1 |= (unsigned long)ttm
551 << (MMCR1_TD_CP_DBG0SEL_SH - 2 * byte);
547 } 552 }
548 553
549 /* Second pass: assign PMCs, set PMCxSEL and PMCx_ADDER_SEL fields */ 554 /* Second pass: assign PMCs, set PMCxSEL and PMCx_ADDER_SEL fields */
@@ -568,7 +573,7 @@ static int power5p_compute_mmcr(u64 event[], int n_ev,
568 if (isbus && (byte & 2) && 573 if (isbus && (byte & 2) &&
569 (psel == 8 || psel == 0x10 || psel == 0x28)) 574 (psel == 8 || psel == 0x10 || psel == 0x28))
570 /* add events on higher-numbered bus */ 575 /* add events on higher-numbered bus */
571 mmcr1 |= 1ull << (MMCR1_PMC1_ADDER_SEL_SH - pmc); 576 mmcr1 |= 1ul << (MMCR1_PMC1_ADDER_SEL_SH - pmc);
572 } else { 577 } else {
573 /* Instructions or run cycles on PMC5/6 */ 578 /* Instructions or run cycles on PMC5/6 */
574 --pmc; 579 --pmc;
@@ -576,7 +581,7 @@ static int power5p_compute_mmcr(u64 event[], int n_ev,
576 if (isbus && unit == PM_GRS) { 581 if (isbus && unit == PM_GRS) {
577 bit = psel & 7; 582 bit = psel & 7;
578 grsel = (event[i] >> PM_GRS_SH) & PM_GRS_MSK; 583 grsel = (event[i] >> PM_GRS_SH) & PM_GRS_MSK;
579 mmcr1 |= (u64)grsel << grsel_shift[bit]; 584 mmcr1 |= (unsigned long)grsel << grsel_shift[bit];
580 } 585 }
581 if (power5p_marked_instr_event(event[i])) 586 if (power5p_marked_instr_event(event[i]))
582 mmcra |= MMCRA_SAMPLE_ENABLE; 587 mmcra |= MMCRA_SAMPLE_ENABLE;
@@ -599,7 +604,7 @@ static int power5p_compute_mmcr(u64 event[], int n_ev,
599 return 0; 604 return 0;
600} 605}
601 606
602static void power5p_disable_pmc(unsigned int pmc, u64 mmcr[]) 607static void power5p_disable_pmc(unsigned int pmc, unsigned long mmcr[])
603{ 608{
604 if (pmc <= 3) 609 if (pmc <= 3)
605 mmcr[1] &= ~(0x7fUL << MMCR1_PMCSEL_SH(pmc)); 610 mmcr[1] &= ~(0x7fUL << MMCR1_PMCSEL_SH(pmc));
@@ -654,18 +659,30 @@ static int power5p_cache_events[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = {
654 }, 659 },
655}; 660};
656 661
657struct power_pmu power5p_pmu = { 662static struct power_pmu power5p_pmu = {
658 .n_counter = 6, 663 .name = "POWER5+/++",
659 .max_alternatives = MAX_ALT, 664 .n_counter = 6,
660 .add_fields = 0x7000000000055ull, 665 .max_alternatives = MAX_ALT,
661 .test_adder = 0x3000040000000ull, 666 .add_fields = 0x7000000000055ul,
662 .compute_mmcr = power5p_compute_mmcr, 667 .test_adder = 0x3000040000000ul,
663 .get_constraint = power5p_get_constraint, 668 .compute_mmcr = power5p_compute_mmcr,
664 .get_alternatives = power5p_get_alternatives, 669 .get_constraint = power5p_get_constraint,
665 .disable_pmc = power5p_disable_pmc, 670 .get_alternatives = power5p_get_alternatives,
666 .limited_pmc_event = power5p_limited_pmc_event, 671 .disable_pmc = power5p_disable_pmc,
667 .flags = PPMU_LIMITED_PMC5_6, 672 .limited_pmc_event = power5p_limited_pmc_event,
668 .n_generic = ARRAY_SIZE(power5p_generic_events), 673 .flags = PPMU_LIMITED_PMC5_6,
669 .generic_events = power5p_generic_events, 674 .n_generic = ARRAY_SIZE(power5p_generic_events),
670 .cache_events = &power5p_cache_events, 675 .generic_events = power5p_generic_events,
676 .cache_events = &power5p_cache_events,
671}; 677};
678
679static int init_power5p_pmu(void)
680{
681 if (strcmp(cur_cpu_spec->oprofile_cpu_type, "ppc64/power5+")
682 && strcmp(cur_cpu_spec->oprofile_cpu_type, "ppc64/power5++"))
683 return -ENODEV;
684
685 return register_power_pmu(&power5p_pmu);
686}
687
688arch_initcall(init_power5p_pmu);
diff --git a/arch/powerpc/kernel/power5-pmu.c b/arch/powerpc/kernel/power5-pmu.c
index 05600b66221a..29b2c6c0e83a 100644
--- a/arch/powerpc/kernel/power5-pmu.c
+++ b/arch/powerpc/kernel/power5-pmu.c
@@ -10,7 +10,9 @@
10 */ 10 */
11#include <linux/kernel.h> 11#include <linux/kernel.h>
12#include <linux/perf_counter.h> 12#include <linux/perf_counter.h>
13#include <linux/string.h>
13#include <asm/reg.h> 14#include <asm/reg.h>
15#include <asm/cputable.h>
14 16
15/* 17/*
16 * Bits in event code for POWER5 (not POWER5++) 18 * Bits in event code for POWER5 (not POWER5++)
@@ -130,20 +132,21 @@ static const int grsel_shift[8] = {
130}; 132};
131 133
132/* Masks and values for using events from the various units */ 134/* Masks and values for using events from the various units */
133static u64 unit_cons[PM_LASTUNIT+1][2] = { 135static unsigned long unit_cons[PM_LASTUNIT+1][2] = {
134 [PM_FPU] = { 0xc0002000000000ull, 0x00001000000000ull }, 136 [PM_FPU] = { 0xc0002000000000ul, 0x00001000000000ul },
135 [PM_ISU0] = { 0x00002000000000ull, 0x00000800000000ull }, 137 [PM_ISU0] = { 0x00002000000000ul, 0x00000800000000ul },
136 [PM_ISU1] = { 0xc0002000000000ull, 0xc0001000000000ull }, 138 [PM_ISU1] = { 0xc0002000000000ul, 0xc0001000000000ul },
137 [PM_IFU] = { 0xc0002000000000ull, 0x80001000000000ull }, 139 [PM_IFU] = { 0xc0002000000000ul, 0x80001000000000ul },
138 [PM_IDU] = { 0x30002000000000ull, 0x00000400000000ull }, 140 [PM_IDU] = { 0x30002000000000ul, 0x00000400000000ul },
139 [PM_GRS] = { 0x30002000000000ull, 0x30000400000000ull }, 141 [PM_GRS] = { 0x30002000000000ul, 0x30000400000000ul },
140}; 142};
141 143
142static int power5_get_constraint(u64 event, u64 *maskp, u64 *valp) 144static int power5_get_constraint(u64 event, unsigned long *maskp,
145 unsigned long *valp)
143{ 146{
144 int pmc, byte, unit, sh; 147 int pmc, byte, unit, sh;
145 int bit, fmask; 148 int bit, fmask;
146 u64 mask = 0, value = 0; 149 unsigned long mask = 0, value = 0;
147 int grp = -1; 150 int grp = -1;
148 151
149 pmc = (event >> PM_PMC_SH) & PM_PMC_MSK; 152 pmc = (event >> PM_PMC_SH) & PM_PMC_MSK;
@@ -178,8 +181,9 @@ static int power5_get_constraint(u64 event, u64 *maskp, u64 *valp)
178 bit = event & 7; 181 bit = event & 7;
179 fmask = (bit == 6)? 7: 3; 182 fmask = (bit == 6)? 7: 3;
180 sh = grsel_shift[bit]; 183 sh = grsel_shift[bit];
181 mask |= (u64)fmask << sh; 184 mask |= (unsigned long)fmask << sh;
182 value |= (u64)((event >> PM_GRS_SH) & fmask) << sh; 185 value |= (unsigned long)((event >> PM_GRS_SH) & fmask)
186 << sh;
183 } 187 }
184 /* 188 /*
185 * Bus events on bytes 0 and 2 can be counted 189 * Bus events on bytes 0 and 2 can be counted
@@ -188,22 +192,22 @@ static int power5_get_constraint(u64 event, u64 *maskp, u64 *valp)
188 if (!pmc) 192 if (!pmc)
189 grp = byte & 1; 193 grp = byte & 1;
190 /* Set byte lane select field */ 194 /* Set byte lane select field */
191 mask |= 0xfULL << (24 - 4 * byte); 195 mask |= 0xfUL << (24 - 4 * byte);
192 value |= (u64)unit << (24 - 4 * byte); 196 value |= (unsigned long)unit << (24 - 4 * byte);
193 } 197 }
194 if (grp == 0) { 198 if (grp == 0) {
195 /* increment PMC1/2 field */ 199 /* increment PMC1/2 field */
196 mask |= 0x200000000ull; 200 mask |= 0x200000000ul;
197 value |= 0x080000000ull; 201 value |= 0x080000000ul;
198 } else if (grp == 1) { 202 } else if (grp == 1) {
199 /* increment PMC3/4 field */ 203 /* increment PMC3/4 field */
200 mask |= 0x40000000ull; 204 mask |= 0x40000000ul;
201 value |= 0x10000000ull; 205 value |= 0x10000000ul;
202 } 206 }
203 if (pmc < 5) { 207 if (pmc < 5) {
204 /* need a counter from PMC1-4 set */ 208 /* need a counter from PMC1-4 set */
205 mask |= 0x8000000000000ull; 209 mask |= 0x8000000000000ul;
206 value |= 0x1000000000000ull; 210 value |= 0x1000000000000ul;
207 } 211 }
208 *maskp = mask; 212 *maskp = mask;
209 *valp = value; 213 *valp = value;
@@ -383,10 +387,10 @@ static int power5_marked_instr_event(u64 event)
383} 387}
384 388
385static int power5_compute_mmcr(u64 event[], int n_ev, 389static int power5_compute_mmcr(u64 event[], int n_ev,
386 unsigned int hwc[], u64 mmcr[]) 390 unsigned int hwc[], unsigned long mmcr[])
387{ 391{
388 u64 mmcr1 = 0; 392 unsigned long mmcr1 = 0;
389 u64 mmcra = 0; 393 unsigned long mmcra = 0;
390 unsigned int pmc, unit, byte, psel; 394 unsigned int pmc, unit, byte, psel;
391 unsigned int ttm, grp; 395 unsigned int ttm, grp;
392 int i, isbus, bit, grsel; 396 int i, isbus, bit, grsel;
@@ -457,7 +461,7 @@ static int power5_compute_mmcr(u64 event[], int n_ev,
457 continue; 461 continue;
458 if (ttmuse++) 462 if (ttmuse++)
459 return -1; 463 return -1;
460 mmcr1 |= (u64)i << MMCR1_TTM0SEL_SH; 464 mmcr1 |= (unsigned long)i << MMCR1_TTM0SEL_SH;
461 } 465 }
462 ttmuse = 0; 466 ttmuse = 0;
463 for (; i <= PM_GRS; ++i) { 467 for (; i <= PM_GRS; ++i) {
@@ -465,7 +469,7 @@ static int power5_compute_mmcr(u64 event[], int n_ev,
465 continue; 469 continue;
466 if (ttmuse++) 470 if (ttmuse++)
467 return -1; 471 return -1;
468 mmcr1 |= (u64)(i & 3) << MMCR1_TTM1SEL_SH; 472 mmcr1 |= (unsigned long)(i & 3) << MMCR1_TTM1SEL_SH;
469 } 473 }
470 if (ttmuse > 1) 474 if (ttmuse > 1)
471 return -1; 475 return -1;
@@ -480,10 +484,11 @@ static int power5_compute_mmcr(u64 event[], int n_ev,
480 unit = PM_ISU0_ALT; 484 unit = PM_ISU0_ALT;
481 } else if (unit == PM_LSU1 + 1) { 485 } else if (unit == PM_LSU1 + 1) {
482 /* select lower word of LSU1 for this byte */ 486 /* select lower word of LSU1 for this byte */
483 mmcr1 |= 1ull << (MMCR1_TTM3SEL_SH + 3 - byte); 487 mmcr1 |= 1ul << (MMCR1_TTM3SEL_SH + 3 - byte);
484 } 488 }
485 ttm = unit >> 2; 489 ttm = unit >> 2;
486 mmcr1 |= (u64)ttm << (MMCR1_TD_CP_DBG0SEL_SH - 2 * byte); 490 mmcr1 |= (unsigned long)ttm
491 << (MMCR1_TD_CP_DBG0SEL_SH - 2 * byte);
487 } 492 }
488 493
489 /* Second pass: assign PMCs, set PMCxSEL and PMCx_ADDER_SEL fields */ 494 /* Second pass: assign PMCs, set PMCxSEL and PMCx_ADDER_SEL fields */
@@ -513,7 +518,7 @@ static int power5_compute_mmcr(u64 event[], int n_ev,
513 --pmc; 518 --pmc;
514 if ((psel == 8 || psel == 0x10) && isbus && (byte & 2)) 519 if ((psel == 8 || psel == 0x10) && isbus && (byte & 2))
515 /* add events on higher-numbered bus */ 520 /* add events on higher-numbered bus */
516 mmcr1 |= 1ull << (MMCR1_PMC1_ADDER_SEL_SH - pmc); 521 mmcr1 |= 1ul << (MMCR1_PMC1_ADDER_SEL_SH - pmc);
517 } else { 522 } else {
518 /* Instructions or run cycles on PMC5/6 */ 523 /* Instructions or run cycles on PMC5/6 */
519 --pmc; 524 --pmc;
@@ -521,7 +526,7 @@ static int power5_compute_mmcr(u64 event[], int n_ev,
521 if (isbus && unit == PM_GRS) { 526 if (isbus && unit == PM_GRS) {
522 bit = psel & 7; 527 bit = psel & 7;
523 grsel = (event[i] >> PM_GRS_SH) & PM_GRS_MSK; 528 grsel = (event[i] >> PM_GRS_SH) & PM_GRS_MSK;
524 mmcr1 |= (u64)grsel << grsel_shift[bit]; 529 mmcr1 |= (unsigned long)grsel << grsel_shift[bit];
525 } 530 }
526 if (power5_marked_instr_event(event[i])) 531 if (power5_marked_instr_event(event[i]))
527 mmcra |= MMCRA_SAMPLE_ENABLE; 532 mmcra |= MMCRA_SAMPLE_ENABLE;
@@ -541,7 +546,7 @@ static int power5_compute_mmcr(u64 event[], int n_ev,
541 return 0; 546 return 0;
542} 547}
543 548
544static void power5_disable_pmc(unsigned int pmc, u64 mmcr[]) 549static void power5_disable_pmc(unsigned int pmc, unsigned long mmcr[])
545{ 550{
546 if (pmc <= 3) 551 if (pmc <= 3)
547 mmcr[1] &= ~(0x7fUL << MMCR1_PMCSEL_SH(pmc)); 552 mmcr[1] &= ~(0x7fUL << MMCR1_PMCSEL_SH(pmc));
@@ -596,16 +601,27 @@ static int power5_cache_events[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = {
596 }, 601 },
597}; 602};
598 603
599struct power_pmu power5_pmu = { 604static struct power_pmu power5_pmu = {
600 .n_counter = 6, 605 .name = "POWER5",
601 .max_alternatives = MAX_ALT, 606 .n_counter = 6,
602 .add_fields = 0x7000090000555ull, 607 .max_alternatives = MAX_ALT,
603 .test_adder = 0x3000490000000ull, 608 .add_fields = 0x7000090000555ul,
604 .compute_mmcr = power5_compute_mmcr, 609 .test_adder = 0x3000490000000ul,
605 .get_constraint = power5_get_constraint, 610 .compute_mmcr = power5_compute_mmcr,
606 .get_alternatives = power5_get_alternatives, 611 .get_constraint = power5_get_constraint,
607 .disable_pmc = power5_disable_pmc, 612 .get_alternatives = power5_get_alternatives,
608 .n_generic = ARRAY_SIZE(power5_generic_events), 613 .disable_pmc = power5_disable_pmc,
609 .generic_events = power5_generic_events, 614 .n_generic = ARRAY_SIZE(power5_generic_events),
610 .cache_events = &power5_cache_events, 615 .generic_events = power5_generic_events,
616 .cache_events = &power5_cache_events,
611}; 617};
618
619static int init_power5_pmu(void)
620{
621 if (strcmp(cur_cpu_spec->oprofile_cpu_type, "ppc64/power5"))
622 return -ENODEV;
623
624 return register_power_pmu(&power5_pmu);
625}
626
627arch_initcall(init_power5_pmu);
diff --git a/arch/powerpc/kernel/power6-pmu.c b/arch/powerpc/kernel/power6-pmu.c
index 46f74bebcfd9..09ae5bf5bda7 100644
--- a/arch/powerpc/kernel/power6-pmu.c
+++ b/arch/powerpc/kernel/power6-pmu.c
@@ -10,7 +10,9 @@
10 */ 10 */
11#include <linux/kernel.h> 11#include <linux/kernel.h>
12#include <linux/perf_counter.h> 12#include <linux/perf_counter.h>
13#include <linux/string.h>
13#include <asm/reg.h> 14#include <asm/reg.h>
15#include <asm/cputable.h>
14 16
15/* 17/*
16 * Bits in event code for POWER6 18 * Bits in event code for POWER6
@@ -41,9 +43,9 @@
41#define MMCR1_NESTSEL_SH 45 43#define MMCR1_NESTSEL_SH 45
42#define MMCR1_NESTSEL_MSK 0x7 44#define MMCR1_NESTSEL_MSK 0x7
43#define MMCR1_NESTSEL(m) (((m) >> MMCR1_NESTSEL_SH) & MMCR1_NESTSEL_MSK) 45#define MMCR1_NESTSEL(m) (((m) >> MMCR1_NESTSEL_SH) & MMCR1_NESTSEL_MSK)
44#define MMCR1_PMC1_LLA ((u64)1 << 44) 46#define MMCR1_PMC1_LLA (1ul << 44)
45#define MMCR1_PMC1_LLA_VALUE ((u64)1 << 39) 47#define MMCR1_PMC1_LLA_VALUE (1ul << 39)
46#define MMCR1_PMC1_ADDR_SEL ((u64)1 << 35) 48#define MMCR1_PMC1_ADDR_SEL (1ul << 35)
47#define MMCR1_PMC1SEL_SH 24 49#define MMCR1_PMC1SEL_SH 24
48#define MMCR1_PMCSEL_SH(n) (MMCR1_PMC1SEL_SH - (n) * 8) 50#define MMCR1_PMCSEL_SH(n) (MMCR1_PMC1SEL_SH - (n) * 8)
49#define MMCR1_PMCSEL_MSK 0xff 51#define MMCR1_PMCSEL_MSK 0xff
@@ -173,10 +175,10 @@ static int power6_marked_instr_event(u64 event)
173 * Assign PMC numbers and compute MMCR1 value for a set of events 175 * Assign PMC numbers and compute MMCR1 value for a set of events
174 */ 176 */
175static int p6_compute_mmcr(u64 event[], int n_ev, 177static int p6_compute_mmcr(u64 event[], int n_ev,
176 unsigned int hwc[], u64 mmcr[]) 178 unsigned int hwc[], unsigned long mmcr[])
177{ 179{
178 u64 mmcr1 = 0; 180 unsigned long mmcr1 = 0;
179 u64 mmcra = 0; 181 unsigned long mmcra = 0;
180 int i; 182 int i;
181 unsigned int pmc, ev, b, u, s, psel; 183 unsigned int pmc, ev, b, u, s, psel;
182 unsigned int ttmset = 0; 184 unsigned int ttmset = 0;
@@ -215,7 +217,7 @@ static int p6_compute_mmcr(u64 event[], int n_ev,
215 /* check for conflict on this byte of event bus */ 217 /* check for conflict on this byte of event bus */
216 if ((ttmset & (1 << b)) && MMCR1_TTMSEL(mmcr1, b) != u) 218 if ((ttmset & (1 << b)) && MMCR1_TTMSEL(mmcr1, b) != u)
217 return -1; 219 return -1;
218 mmcr1 |= (u64)u << MMCR1_TTMSEL_SH(b); 220 mmcr1 |= (unsigned long)u << MMCR1_TTMSEL_SH(b);
219 ttmset |= 1 << b; 221 ttmset |= 1 << b;
220 if (u == 5) { 222 if (u == 5) {
221 /* Nest events have a further mux */ 223 /* Nest events have a further mux */
@@ -224,7 +226,7 @@ static int p6_compute_mmcr(u64 event[], int n_ev,
224 MMCR1_NESTSEL(mmcr1) != s) 226 MMCR1_NESTSEL(mmcr1) != s)
225 return -1; 227 return -1;
226 ttmset |= 0x10; 228 ttmset |= 0x10;
227 mmcr1 |= (u64)s << MMCR1_NESTSEL_SH; 229 mmcr1 |= (unsigned long)s << MMCR1_NESTSEL_SH;
228 } 230 }
229 if (0x30 <= psel && psel <= 0x3d) { 231 if (0x30 <= psel && psel <= 0x3d) {
230 /* these need the PMCx_ADDR_SEL bits */ 232 /* these need the PMCx_ADDR_SEL bits */
@@ -243,7 +245,7 @@ static int p6_compute_mmcr(u64 event[], int n_ev,
243 if (power6_marked_instr_event(event[i])) 245 if (power6_marked_instr_event(event[i]))
244 mmcra |= MMCRA_SAMPLE_ENABLE; 246 mmcra |= MMCRA_SAMPLE_ENABLE;
245 if (pmc < 4) 247 if (pmc < 4)
246 mmcr1 |= (u64)psel << MMCR1_PMCSEL_SH(pmc); 248 mmcr1 |= (unsigned long)psel << MMCR1_PMCSEL_SH(pmc);
247 } 249 }
248 mmcr[0] = 0; 250 mmcr[0] = 0;
249 if (pmc_inuse & 1) 251 if (pmc_inuse & 1)
@@ -265,10 +267,11 @@ static int p6_compute_mmcr(u64 event[], int n_ev,
265 * 20-23, 24-27, 28-31 ditto for bytes 1, 2, 3 267 * 20-23, 24-27, 28-31 ditto for bytes 1, 2, 3
266 * 32-34 select field: nest (subunit) event selector 268 * 32-34 select field: nest (subunit) event selector
267 */ 269 */
268static int p6_get_constraint(u64 event, u64 *maskp, u64 *valp) 270static int p6_get_constraint(u64 event, unsigned long *maskp,
271 unsigned long *valp)
269{ 272{
270 int pmc, byte, sh, subunit; 273 int pmc, byte, sh, subunit;
271 u64 mask = 0, value = 0; 274 unsigned long mask = 0, value = 0;
272 275
273 pmc = (event >> PM_PMC_SH) & PM_PMC_MSK; 276 pmc = (event >> PM_PMC_SH) & PM_PMC_MSK;
274 if (pmc) { 277 if (pmc) {
@@ -282,11 +285,11 @@ static int p6_get_constraint(u64 event, u64 *maskp, u64 *valp)
282 byte = (event >> PM_BYTE_SH) & PM_BYTE_MSK; 285 byte = (event >> PM_BYTE_SH) & PM_BYTE_MSK;
283 sh = byte * 4 + (16 - PM_UNIT_SH); 286 sh = byte * 4 + (16 - PM_UNIT_SH);
284 mask |= PM_UNIT_MSKS << sh; 287 mask |= PM_UNIT_MSKS << sh;
285 value |= (u64)(event & PM_UNIT_MSKS) << sh; 288 value |= (unsigned long)(event & PM_UNIT_MSKS) << sh;
286 if ((event & PM_UNIT_MSKS) == (5 << PM_UNIT_SH)) { 289 if ((event & PM_UNIT_MSKS) == (5 << PM_UNIT_SH)) {
287 subunit = (event >> PM_SUBUNIT_SH) & PM_SUBUNIT_MSK; 290 subunit = (event >> PM_SUBUNIT_SH) & PM_SUBUNIT_MSK;
288 mask |= (u64)PM_SUBUNIT_MSK << 32; 291 mask |= (unsigned long)PM_SUBUNIT_MSK << 32;
289 value |= (u64)subunit << 32; 292 value |= (unsigned long)subunit << 32;
290 } 293 }
291 } 294 }
292 if (pmc <= 4) { 295 if (pmc <= 4) {
@@ -458,7 +461,7 @@ static int p6_get_alternatives(u64 event, unsigned int flags, u64 alt[])
458 return nalt; 461 return nalt;
459} 462}
460 463
461static void p6_disable_pmc(unsigned int pmc, u64 mmcr[]) 464static void p6_disable_pmc(unsigned int pmc, unsigned long mmcr[])
462{ 465{
463 /* Set PMCxSEL to 0 to disable PMCx */ 466 /* Set PMCxSEL to 0 to disable PMCx */
464 if (pmc <= 3) 467 if (pmc <= 3)
@@ -515,18 +518,29 @@ static int power6_cache_events[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = {
515 }, 518 },
516}; 519};
517 520
518struct power_pmu power6_pmu = { 521static struct power_pmu power6_pmu = {
519 .n_counter = 6, 522 .name = "POWER6",
520 .max_alternatives = MAX_ALT, 523 .n_counter = 6,
521 .add_fields = 0x1555, 524 .max_alternatives = MAX_ALT,
522 .test_adder = 0x3000, 525 .add_fields = 0x1555,
523 .compute_mmcr = p6_compute_mmcr, 526 .test_adder = 0x3000,
524 .get_constraint = p6_get_constraint, 527 .compute_mmcr = p6_compute_mmcr,
525 .get_alternatives = p6_get_alternatives, 528 .get_constraint = p6_get_constraint,
526 .disable_pmc = p6_disable_pmc, 529 .get_alternatives = p6_get_alternatives,
527 .limited_pmc_event = p6_limited_pmc_event, 530 .disable_pmc = p6_disable_pmc,
528 .flags = PPMU_LIMITED_PMC5_6 | PPMU_ALT_SIPR, 531 .limited_pmc_event = p6_limited_pmc_event,
529 .n_generic = ARRAY_SIZE(power6_generic_events), 532 .flags = PPMU_LIMITED_PMC5_6 | PPMU_ALT_SIPR,
530 .generic_events = power6_generic_events, 533 .n_generic = ARRAY_SIZE(power6_generic_events),
531 .cache_events = &power6_cache_events, 534 .generic_events = power6_generic_events,
535 .cache_events = &power6_cache_events,
532}; 536};
537
538static int init_power6_pmu(void)
539{
540 if (strcmp(cur_cpu_spec->oprofile_cpu_type, "ppc64/power6"))
541 return -ENODEV;
542
543 return register_power_pmu(&power6_pmu);
544}
545
546arch_initcall(init_power6_pmu);
diff --git a/arch/powerpc/kernel/power7-pmu.c b/arch/powerpc/kernel/power7-pmu.c
index b72e7a19d054..5d755ef7ac8f 100644
--- a/arch/powerpc/kernel/power7-pmu.c
+++ b/arch/powerpc/kernel/power7-pmu.c
@@ -10,7 +10,9 @@
10 */ 10 */
11#include <linux/kernel.h> 11#include <linux/kernel.h>
12#include <linux/perf_counter.h> 12#include <linux/perf_counter.h>
13#include <linux/string.h>
13#include <asm/reg.h> 14#include <asm/reg.h>
15#include <asm/cputable.h>
14 16
15/* 17/*
16 * Bits in event code for POWER7 18 * Bits in event code for POWER7
@@ -71,10 +73,11 @@
71 * 0-9: Count of events needing PMC1..PMC5 73 * 0-9: Count of events needing PMC1..PMC5
72 */ 74 */
73 75
74static int power7_get_constraint(u64 event, u64 *maskp, u64 *valp) 76static int power7_get_constraint(u64 event, unsigned long *maskp,
77 unsigned long *valp)
75{ 78{
76 int pmc, sh; 79 int pmc, sh;
77 u64 mask = 0, value = 0; 80 unsigned long mask = 0, value = 0;
78 81
79 pmc = (event >> PM_PMC_SH) & PM_PMC_MSK; 82 pmc = (event >> PM_PMC_SH) & PM_PMC_MSK;
80 if (pmc) { 83 if (pmc) {
@@ -224,10 +227,10 @@ static int power7_marked_instr_event(u64 event)
224} 227}
225 228
226static int power7_compute_mmcr(u64 event[], int n_ev, 229static int power7_compute_mmcr(u64 event[], int n_ev,
227 unsigned int hwc[], u64 mmcr[]) 230 unsigned int hwc[], unsigned long mmcr[])
228{ 231{
229 u64 mmcr1 = 0; 232 unsigned long mmcr1 = 0;
230 u64 mmcra = 0; 233 unsigned long mmcra = 0;
231 unsigned int pmc, unit, combine, l2sel, psel; 234 unsigned int pmc, unit, combine, l2sel, psel;
232 unsigned int pmc_inuse = 0; 235 unsigned int pmc_inuse = 0;
233 int i; 236 int i;
@@ -265,11 +268,14 @@ static int power7_compute_mmcr(u64 event[], int n_ev,
265 --pmc; 268 --pmc;
266 } 269 }
267 if (pmc <= 3) { 270 if (pmc <= 3) {
268 mmcr1 |= (u64) unit << (MMCR1_TTM0SEL_SH - 4 * pmc); 271 mmcr1 |= (unsigned long) unit
269 mmcr1 |= (u64) combine << (MMCR1_PMC1_COMBINE_SH - pmc); 272 << (MMCR1_TTM0SEL_SH - 4 * pmc);
273 mmcr1 |= (unsigned long) combine
274 << (MMCR1_PMC1_COMBINE_SH - pmc);
270 mmcr1 |= psel << MMCR1_PMCSEL_SH(pmc); 275 mmcr1 |= psel << MMCR1_PMCSEL_SH(pmc);
271 if (unit == 6) /* L2 events */ 276 if (unit == 6) /* L2 events */
272 mmcr1 |= (u64) l2sel << MMCR1_L2SEL_SH; 277 mmcr1 |= (unsigned long) l2sel
278 << MMCR1_L2SEL_SH;
273 } 279 }
274 if (power7_marked_instr_event(event[i])) 280 if (power7_marked_instr_event(event[i]))
275 mmcra |= MMCRA_SAMPLE_ENABLE; 281 mmcra |= MMCRA_SAMPLE_ENABLE;
@@ -287,10 +293,10 @@ static int power7_compute_mmcr(u64 event[], int n_ev,
287 return 0; 293 return 0;
288} 294}
289 295
290static void power7_disable_pmc(unsigned int pmc, u64 mmcr[]) 296static void power7_disable_pmc(unsigned int pmc, unsigned long mmcr[])
291{ 297{
292 if (pmc <= 3) 298 if (pmc <= 3)
293 mmcr[1] &= ~(0xffULL << MMCR1_PMCSEL_SH(pmc)); 299 mmcr[1] &= ~(0xffUL << MMCR1_PMCSEL_SH(pmc));
294} 300}
295 301
296static int power7_generic_events[] = { 302static int power7_generic_events[] = {
@@ -342,16 +348,27 @@ static int power7_cache_events[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = {
342 }, 348 },
343}; 349};
344 350
345struct power_pmu power7_pmu = { 351static struct power_pmu power7_pmu = {
346 .n_counter = 6, 352 .name = "POWER7",
347 .max_alternatives = MAX_ALT + 1, 353 .n_counter = 6,
348 .add_fields = 0x1555ull, 354 .max_alternatives = MAX_ALT + 1,
349 .test_adder = 0x3000ull, 355 .add_fields = 0x1555ul,
350 .compute_mmcr = power7_compute_mmcr, 356 .test_adder = 0x3000ul,
351 .get_constraint = power7_get_constraint, 357 .compute_mmcr = power7_compute_mmcr,
352 .get_alternatives = power7_get_alternatives, 358 .get_constraint = power7_get_constraint,
353 .disable_pmc = power7_disable_pmc, 359 .get_alternatives = power7_get_alternatives,
354 .n_generic = ARRAY_SIZE(power7_generic_events), 360 .disable_pmc = power7_disable_pmc,
355 .generic_events = power7_generic_events, 361 .n_generic = ARRAY_SIZE(power7_generic_events),
356 .cache_events = &power7_cache_events, 362 .generic_events = power7_generic_events,
363 .cache_events = &power7_cache_events,
357}; 364};
365
366static int init_power7_pmu(void)
367{
368 if (strcmp(cur_cpu_spec->oprofile_cpu_type, "ppc64/power7"))
369 return -ENODEV;
370
371 return register_power_pmu(&power7_pmu);
372}
373
374arch_initcall(init_power7_pmu);
diff --git a/arch/powerpc/kernel/ppc970-pmu.c b/arch/powerpc/kernel/ppc970-pmu.c
index ba0a357a89f4..6637c87fe70e 100644
--- a/arch/powerpc/kernel/ppc970-pmu.c
+++ b/arch/powerpc/kernel/ppc970-pmu.c
@@ -10,7 +10,9 @@
10 */ 10 */
11#include <linux/string.h> 11#include <linux/string.h>
12#include <linux/perf_counter.h> 12#include <linux/perf_counter.h>
13#include <linux/string.h>
13#include <asm/reg.h> 14#include <asm/reg.h>
15#include <asm/cputable.h>
14 16
15/* 17/*
16 * Bits in event code for PPC970 18 * Bits in event code for PPC970
@@ -183,7 +185,7 @@ static int p970_marked_instr_event(u64 event)
183} 185}
184 186
185/* Masks and values for using events from the various units */ 187/* Masks and values for using events from the various units */
186static u64 unit_cons[PM_LASTUNIT+1][2] = { 188static unsigned long unit_cons[PM_LASTUNIT+1][2] = {
187 [PM_FPU] = { 0xc80000000000ull, 0x040000000000ull }, 189 [PM_FPU] = { 0xc80000000000ull, 0x040000000000ull },
188 [PM_VPU] = { 0xc80000000000ull, 0xc40000000000ull }, 190 [PM_VPU] = { 0xc80000000000ull, 0xc40000000000ull },
189 [PM_ISU] = { 0x080000000000ull, 0x020000000000ull }, 191 [PM_ISU] = { 0x080000000000ull, 0x020000000000ull },
@@ -192,10 +194,11 @@ static u64 unit_cons[PM_LASTUNIT+1][2] = {
192 [PM_STS] = { 0x380000000000ull, 0x310000000000ull }, 194 [PM_STS] = { 0x380000000000ull, 0x310000000000ull },
193}; 195};
194 196
195static int p970_get_constraint(u64 event, u64 *maskp, u64 *valp) 197static int p970_get_constraint(u64 event, unsigned long *maskp,
198 unsigned long *valp)
196{ 199{
197 int pmc, byte, unit, sh, spcsel; 200 int pmc, byte, unit, sh, spcsel;
198 u64 mask = 0, value = 0; 201 unsigned long mask = 0, value = 0;
199 int grp = -1; 202 int grp = -1;
200 203
201 pmc = (event >> PM_PMC_SH) & PM_PMC_MSK; 204 pmc = (event >> PM_PMC_SH) & PM_PMC_MSK;
@@ -222,7 +225,7 @@ static int p970_get_constraint(u64 event, u64 *maskp, u64 *valp)
222 grp = byte & 1; 225 grp = byte & 1;
223 /* Set byte lane select field */ 226 /* Set byte lane select field */
224 mask |= 0xfULL << (28 - 4 * byte); 227 mask |= 0xfULL << (28 - 4 * byte);
225 value |= (u64)unit << (28 - 4 * byte); 228 value |= (unsigned long)unit << (28 - 4 * byte);
226 } 229 }
227 if (grp == 0) { 230 if (grp == 0) {
228 /* increment PMC1/2/5/6 field */ 231 /* increment PMC1/2/5/6 field */
@@ -236,7 +239,7 @@ static int p970_get_constraint(u64 event, u64 *maskp, u64 *valp)
236 spcsel = (event >> PM_SPCSEL_SH) & PM_SPCSEL_MSK; 239 spcsel = (event >> PM_SPCSEL_SH) & PM_SPCSEL_MSK;
237 if (spcsel) { 240 if (spcsel) {
238 mask |= 3ull << 48; 241 mask |= 3ull << 48;
239 value |= (u64)spcsel << 48; 242 value |= (unsigned long)spcsel << 48;
240 } 243 }
241 *maskp = mask; 244 *maskp = mask;
242 *valp = value; 245 *valp = value;
@@ -257,9 +260,9 @@ static int p970_get_alternatives(u64 event, unsigned int flags, u64 alt[])
257} 260}
258 261
259static int p970_compute_mmcr(u64 event[], int n_ev, 262static int p970_compute_mmcr(u64 event[], int n_ev,
260 unsigned int hwc[], u64 mmcr[]) 263 unsigned int hwc[], unsigned long mmcr[])
261{ 264{
262 u64 mmcr0 = 0, mmcr1 = 0, mmcra = 0; 265 unsigned long mmcr0 = 0, mmcr1 = 0, mmcra = 0;
263 unsigned int pmc, unit, byte, psel; 266 unsigned int pmc, unit, byte, psel;
264 unsigned int ttm, grp; 267 unsigned int ttm, grp;
265 unsigned int pmc_inuse = 0; 268 unsigned int pmc_inuse = 0;
@@ -320,7 +323,7 @@ static int p970_compute_mmcr(u64 event[], int n_ev,
320 continue; 323 continue;
321 ttm = unitmap[i]; 324 ttm = unitmap[i];
322 ++ttmuse[(ttm >> 2) & 1]; 325 ++ttmuse[(ttm >> 2) & 1];
323 mmcr1 |= (u64)(ttm & ~4) << MMCR1_TTM1SEL_SH; 326 mmcr1 |= (unsigned long)(ttm & ~4) << MMCR1_TTM1SEL_SH;
324 } 327 }
325 /* Check only one unit per TTMx */ 328 /* Check only one unit per TTMx */
326 if (ttmuse[0] > 1 || ttmuse[1] > 1) 329 if (ttmuse[0] > 1 || ttmuse[1] > 1)
@@ -340,7 +343,8 @@ static int p970_compute_mmcr(u64 event[], int n_ev,
340 if (unit == PM_LSU1L && byte >= 2) 343 if (unit == PM_LSU1L && byte >= 2)
341 mmcr1 |= 1ull << (MMCR1_TTM3SEL_SH + 3 - byte); 344 mmcr1 |= 1ull << (MMCR1_TTM3SEL_SH + 3 - byte);
342 } 345 }
343 mmcr1 |= (u64)ttm << (MMCR1_TD_CP_DBG0SEL_SH - 2 * byte); 346 mmcr1 |= (unsigned long)ttm
347 << (MMCR1_TD_CP_DBG0SEL_SH - 2 * byte);
344 } 348 }
345 349
346 /* Second pass: assign PMCs, set PMCxSEL and PMCx_ADDER_SEL fields */ 350 /* Second pass: assign PMCs, set PMCxSEL and PMCx_ADDER_SEL fields */
@@ -386,7 +390,8 @@ static int p970_compute_mmcr(u64 event[], int n_ev,
386 for (pmc = 0; pmc < 2; ++pmc) 390 for (pmc = 0; pmc < 2; ++pmc)
387 mmcr0 |= pmcsel[pmc] << (MMCR0_PMC1SEL_SH - 7 * pmc); 391 mmcr0 |= pmcsel[pmc] << (MMCR0_PMC1SEL_SH - 7 * pmc);
388 for (; pmc < 8; ++pmc) 392 for (; pmc < 8; ++pmc)
389 mmcr1 |= (u64)pmcsel[pmc] << (MMCR1_PMC3SEL_SH - 5 * (pmc - 2)); 393 mmcr1 |= (unsigned long)pmcsel[pmc]
394 << (MMCR1_PMC3SEL_SH - 5 * (pmc - 2));
390 if (pmc_inuse & 1) 395 if (pmc_inuse & 1)
391 mmcr0 |= MMCR0_PMC1CE; 396 mmcr0 |= MMCR0_PMC1CE;
392 if (pmc_inuse & 0xfe) 397 if (pmc_inuse & 0xfe)
@@ -401,7 +406,7 @@ static int p970_compute_mmcr(u64 event[], int n_ev,
401 return 0; 406 return 0;
402} 407}
403 408
404static void p970_disable_pmc(unsigned int pmc, u64 mmcr[]) 409static void p970_disable_pmc(unsigned int pmc, unsigned long mmcr[])
405{ 410{
406 int shift, i; 411 int shift, i;
407 412
@@ -467,16 +472,28 @@ static int ppc970_cache_events[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = {
467 }, 472 },
468}; 473};
469 474
470struct power_pmu ppc970_pmu = { 475static struct power_pmu ppc970_pmu = {
471 .n_counter = 8, 476 .name = "PPC970/FX/MP",
472 .max_alternatives = 2, 477 .n_counter = 8,
473 .add_fields = 0x001100005555ull, 478 .max_alternatives = 2,
474 .test_adder = 0x013300000000ull, 479 .add_fields = 0x001100005555ull,
475 .compute_mmcr = p970_compute_mmcr, 480 .test_adder = 0x013300000000ull,
476 .get_constraint = p970_get_constraint, 481 .compute_mmcr = p970_compute_mmcr,
477 .get_alternatives = p970_get_alternatives, 482 .get_constraint = p970_get_constraint,
478 .disable_pmc = p970_disable_pmc, 483 .get_alternatives = p970_get_alternatives,
479 .n_generic = ARRAY_SIZE(ppc970_generic_events), 484 .disable_pmc = p970_disable_pmc,
480 .generic_events = ppc970_generic_events, 485 .n_generic = ARRAY_SIZE(ppc970_generic_events),
481 .cache_events = &ppc970_cache_events, 486 .generic_events = ppc970_generic_events,
487 .cache_events = &ppc970_cache_events,
482}; 488};
489
490static int init_ppc970_pmu(void)
491{
492 if (strcmp(cur_cpu_spec->oprofile_cpu_type, "ppc64/970")
493 && strcmp(cur_cpu_spec->oprofile_cpu_type, "ppc64/970MP"))
494 return -ENODEV;
495
496 return register_power_pmu(&ppc970_pmu);
497}
498
499arch_initcall(init_ppc970_pmu);
diff --git a/arch/powerpc/kernel/time.c b/arch/powerpc/kernel/time.c
index 15391c2ab013..eae4511ceeac 100644
--- a/arch/powerpc/kernel/time.c
+++ b/arch/powerpc/kernel/time.c
@@ -53,6 +53,7 @@
53#include <linux/posix-timers.h> 53#include <linux/posix-timers.h>
54#include <linux/irq.h> 54#include <linux/irq.h>
55#include <linux/delay.h> 55#include <linux/delay.h>
56#include <linux/perf_counter.h>
56 57
57#include <asm/io.h> 58#include <asm/io.h>
58#include <asm/processor.h> 59#include <asm/processor.h>
@@ -525,6 +526,26 @@ void __init iSeries_time_init_early(void)
525} 526}
526#endif /* CONFIG_PPC_ISERIES */ 527#endif /* CONFIG_PPC_ISERIES */
527 528
529#if defined(CONFIG_PERF_COUNTERS) && defined(CONFIG_PPC32)
530DEFINE_PER_CPU(u8, perf_counter_pending);
531
532void set_perf_counter_pending(void)
533{
534 get_cpu_var(perf_counter_pending) = 1;
535 set_dec(1);
536 put_cpu_var(perf_counter_pending);
537}
538
539#define test_perf_counter_pending() __get_cpu_var(perf_counter_pending)
540#define clear_perf_counter_pending() __get_cpu_var(perf_counter_pending) = 0
541
542#else /* CONFIG_PERF_COUNTERS && CONFIG_PPC32 */
543
544#define test_perf_counter_pending() 0
545#define clear_perf_counter_pending()
546
547#endif /* CONFIG_PERF_COUNTERS && CONFIG_PPC32 */
548
528/* 549/*
529 * For iSeries shared processors, we have to let the hypervisor 550 * For iSeries shared processors, we have to let the hypervisor
530 * set the hardware decrementer. We set a virtual decrementer 551 * set the hardware decrementer. We set a virtual decrementer
@@ -551,6 +572,10 @@ void timer_interrupt(struct pt_regs * regs)
551 set_dec(DECREMENTER_MAX); 572 set_dec(DECREMENTER_MAX);
552 573
553#ifdef CONFIG_PPC32 574#ifdef CONFIG_PPC32
575 if (test_perf_counter_pending()) {
576 clear_perf_counter_pending();
577 perf_counter_do_pending();
578 }
554 if (atomic_read(&ppc_n_lost_interrupts) != 0) 579 if (atomic_read(&ppc_n_lost_interrupts) != 0)
555 do_IRQ(regs); 580 do_IRQ(regs);
556#endif 581#endif
diff --git a/arch/powerpc/platforms/Kconfig.cputype b/arch/powerpc/platforms/Kconfig.cputype
index c4192542b809..61187bec7506 100644
--- a/arch/powerpc/platforms/Kconfig.cputype
+++ b/arch/powerpc/platforms/Kconfig.cputype
@@ -1,7 +1,7 @@
1config PPC64 1config PPC64
2 bool "64-bit kernel" 2 bool "64-bit kernel"
3 default n 3 default n
4 select HAVE_PERF_COUNTERS 4 select PPC_HAVE_PMU_SUPPORT
5 help 5 help
6 This option selects whether a 32-bit or a 64-bit kernel 6 This option selects whether a 32-bit or a 64-bit kernel
7 will be built. 7 will be built.
@@ -78,6 +78,7 @@ config POWER4_ONLY
78config 6xx 78config 6xx
79 def_bool y 79 def_bool y
80 depends on PPC32 && PPC_BOOK3S 80 depends on PPC32 && PPC_BOOK3S
81 select PPC_HAVE_PMU_SUPPORT
81 82
82config POWER3 83config POWER3
83 bool 84 bool
@@ -246,6 +247,15 @@ config VIRT_CPU_ACCOUNTING
246 247
247 If in doubt, say Y here. 248 If in doubt, say Y here.
248 249
250config PPC_HAVE_PMU_SUPPORT
251 bool
252
253config PPC_PERF_CTRS
254 def_bool y
255 depends on PERF_COUNTERS && PPC_HAVE_PMU_SUPPORT
256 help
257 This enables the powerpc-specific perf_counter back-end.
258
249config SMP 259config SMP
250 depends on PPC_STD_MMU || FSL_BOOKE 260 depends on PPC_STD_MMU || FSL_BOOKE
251 bool "Symmetric multi-processing support" 261 bool "Symmetric multi-processing support"
diff --git a/arch/x86/include/asm/perf_counter.h b/arch/x86/include/asm/perf_counter.h
index 876ed97147b3..5fb33e160ea0 100644
--- a/arch/x86/include/asm/perf_counter.h
+++ b/arch/x86/include/asm/perf_counter.h
@@ -84,11 +84,6 @@ union cpuid10_edx {
84#define MSR_ARCH_PERFMON_FIXED_CTR2 0x30b 84#define MSR_ARCH_PERFMON_FIXED_CTR2 0x30b
85#define X86_PMC_IDX_FIXED_BUS_CYCLES (X86_PMC_IDX_FIXED + 2) 85#define X86_PMC_IDX_FIXED_BUS_CYCLES (X86_PMC_IDX_FIXED + 2)
86 86
87extern void set_perf_counter_pending(void);
88
89#define clear_perf_counter_pending() do { } while (0)
90#define test_perf_counter_pending() (0)
91
92#ifdef CONFIG_PERF_COUNTERS 87#ifdef CONFIG_PERF_COUNTERS
93extern void init_hw_perf_counters(void); 88extern void init_hw_perf_counters(void);
94extern void perf_counters_lapic_init(void); 89extern void perf_counters_lapic_init(void);
diff --git a/arch/x86/include/asm/pgtable_32.h b/arch/x86/include/asm/pgtable_32.h
index 31bd120cf2a2..01fd9461d323 100644
--- a/arch/x86/include/asm/pgtable_32.h
+++ b/arch/x86/include/asm/pgtable_32.h
@@ -49,13 +49,17 @@ extern void set_pmd_pfn(unsigned long, unsigned long, pgprot_t);
49#endif 49#endif
50 50
51#if defined(CONFIG_HIGHPTE) 51#if defined(CONFIG_HIGHPTE)
52#define __KM_PTE \
53 (in_nmi() ? KM_NMI_PTE : \
54 in_irq() ? KM_IRQ_PTE : \
55 KM_PTE0)
52#define pte_offset_map(dir, address) \ 56#define pte_offset_map(dir, address) \
53 ((pte_t *)kmap_atomic_pte(pmd_page(*(dir)), KM_PTE0) + \ 57 ((pte_t *)kmap_atomic_pte(pmd_page(*(dir)), __KM_PTE) + \
54 pte_index((address))) 58 pte_index((address)))
55#define pte_offset_map_nested(dir, address) \ 59#define pte_offset_map_nested(dir, address) \
56 ((pte_t *)kmap_atomic_pte(pmd_page(*(dir)), KM_PTE1) + \ 60 ((pte_t *)kmap_atomic_pte(pmd_page(*(dir)), KM_PTE1) + \
57 pte_index((address))) 61 pte_index((address)))
58#define pte_unmap(pte) kunmap_atomic((pte), KM_PTE0) 62#define pte_unmap(pte) kunmap_atomic((pte), __KM_PTE)
59#define pte_unmap_nested(pte) kunmap_atomic((pte), KM_PTE1) 63#define pte_unmap_nested(pte) kunmap_atomic((pte), KM_PTE1)
60#else 64#else
61#define pte_offset_map(dir, address) \ 65#define pte_offset_map(dir, address) \
diff --git a/arch/x86/include/asm/uaccess.h b/arch/x86/include/asm/uaccess.h
index b685ece89d5c..512ee87062c2 100644
--- a/arch/x86/include/asm/uaccess.h
+++ b/arch/x86/include/asm/uaccess.h
@@ -25,7 +25,12 @@
25#define MAKE_MM_SEG(s) ((mm_segment_t) { (s) }) 25#define MAKE_MM_SEG(s) ((mm_segment_t) { (s) })
26 26
27#define KERNEL_DS MAKE_MM_SEG(-1UL) 27#define KERNEL_DS MAKE_MM_SEG(-1UL)
28#define USER_DS MAKE_MM_SEG(PAGE_OFFSET) 28
29#ifdef CONFIG_X86_32
30# define USER_DS MAKE_MM_SEG(PAGE_OFFSET)
31#else
32# define USER_DS MAKE_MM_SEG(__VIRTUAL_MASK)
33#endif
29 34
30#define get_ds() (KERNEL_DS) 35#define get_ds() (KERNEL_DS)
31#define get_fs() (current_thread_info()->addr_limit) 36#define get_fs() (current_thread_info()->addr_limit)
diff --git a/arch/x86/kernel/cpu/perf_counter.c b/arch/x86/kernel/cpu/perf_counter.c
index 275bc142cd5d..76dfef23f789 100644
--- a/arch/x86/kernel/cpu/perf_counter.c
+++ b/arch/x86/kernel/cpu/perf_counter.c
@@ -19,6 +19,7 @@
19#include <linux/kdebug.h> 19#include <linux/kdebug.h>
20#include <linux/sched.h> 20#include <linux/sched.h>
21#include <linux/uaccess.h> 21#include <linux/uaccess.h>
22#include <linux/highmem.h>
22 23
23#include <asm/apic.h> 24#include <asm/apic.h>
24#include <asm/stacktrace.h> 25#include <asm/stacktrace.h>
@@ -389,23 +390,23 @@ static u64 intel_pmu_raw_event(u64 event)
389 return event & CORE_EVNTSEL_MASK; 390 return event & CORE_EVNTSEL_MASK;
390} 391}
391 392
392static const u64 amd_0f_hw_cache_event_ids 393static const u64 amd_hw_cache_event_ids
393 [PERF_COUNT_HW_CACHE_MAX] 394 [PERF_COUNT_HW_CACHE_MAX]
394 [PERF_COUNT_HW_CACHE_OP_MAX] 395 [PERF_COUNT_HW_CACHE_OP_MAX]
395 [PERF_COUNT_HW_CACHE_RESULT_MAX] = 396 [PERF_COUNT_HW_CACHE_RESULT_MAX] =
396{ 397{
397 [ C(L1D) ] = { 398 [ C(L1D) ] = {
398 [ C(OP_READ) ] = { 399 [ C(OP_READ) ] = {
399 [ C(RESULT_ACCESS) ] = 0, 400 [ C(RESULT_ACCESS) ] = 0x0040, /* Data Cache Accesses */
400 [ C(RESULT_MISS) ] = 0, 401 [ C(RESULT_MISS) ] = 0x0041, /* Data Cache Misses */
401 }, 402 },
402 [ C(OP_WRITE) ] = { 403 [ C(OP_WRITE) ] = {
403 [ C(RESULT_ACCESS) ] = 0, 404 [ C(RESULT_ACCESS) ] = 0x0042, /* Data Cache Refills from L2 */
404 [ C(RESULT_MISS) ] = 0, 405 [ C(RESULT_MISS) ] = 0,
405 }, 406 },
406 [ C(OP_PREFETCH) ] = { 407 [ C(OP_PREFETCH) ] = {
407 [ C(RESULT_ACCESS) ] = 0, 408 [ C(RESULT_ACCESS) ] = 0x0267, /* Data Prefetcher :attempts */
408 [ C(RESULT_MISS) ] = 0, 409 [ C(RESULT_MISS) ] = 0x0167, /* Data Prefetcher :cancelled */
409 }, 410 },
410 }, 411 },
411 [ C(L1I ) ] = { 412 [ C(L1I ) ] = {
@@ -418,17 +419,17 @@ static const u64 amd_0f_hw_cache_event_ids
418 [ C(RESULT_MISS) ] = -1, 419 [ C(RESULT_MISS) ] = -1,
419 }, 420 },
420 [ C(OP_PREFETCH) ] = { 421 [ C(OP_PREFETCH) ] = {
421 [ C(RESULT_ACCESS) ] = 0, 422 [ C(RESULT_ACCESS) ] = 0x014B, /* Prefetch Instructions :Load */
422 [ C(RESULT_MISS) ] = 0, 423 [ C(RESULT_MISS) ] = 0,
423 }, 424 },
424 }, 425 },
425 [ C(LL ) ] = { 426 [ C(LL ) ] = {
426 [ C(OP_READ) ] = { 427 [ C(OP_READ) ] = {
427 [ C(RESULT_ACCESS) ] = 0, 428 [ C(RESULT_ACCESS) ] = 0x037D, /* Requests to L2 Cache :IC+DC */
428 [ C(RESULT_MISS) ] = 0, 429 [ C(RESULT_MISS) ] = 0x037E, /* L2 Cache Misses : IC+DC */
429 }, 430 },
430 [ C(OP_WRITE) ] = { 431 [ C(OP_WRITE) ] = {
431 [ C(RESULT_ACCESS) ] = 0, 432 [ C(RESULT_ACCESS) ] = 0x017F, /* L2 Fill/Writeback */
432 [ C(RESULT_MISS) ] = 0, 433 [ C(RESULT_MISS) ] = 0,
433 }, 434 },
434 [ C(OP_PREFETCH) ] = { 435 [ C(OP_PREFETCH) ] = {
@@ -438,8 +439,8 @@ static const u64 amd_0f_hw_cache_event_ids
438 }, 439 },
439 [ C(DTLB) ] = { 440 [ C(DTLB) ] = {
440 [ C(OP_READ) ] = { 441 [ C(OP_READ) ] = {
441 [ C(RESULT_ACCESS) ] = 0, 442 [ C(RESULT_ACCESS) ] = 0x0040, /* Data Cache Accesses */
442 [ C(RESULT_MISS) ] = 0, 443 [ C(RESULT_MISS) ] = 0x0046, /* L1 DTLB and L2 DLTB Miss */
443 }, 444 },
444 [ C(OP_WRITE) ] = { 445 [ C(OP_WRITE) ] = {
445 [ C(RESULT_ACCESS) ] = 0, 446 [ C(RESULT_ACCESS) ] = 0,
@@ -1223,6 +1224,8 @@ again:
1223 if (!intel_pmu_save_and_restart(counter)) 1224 if (!intel_pmu_save_and_restart(counter))
1224 continue; 1225 continue;
1225 1226
1227 data.period = counter->hw.last_period;
1228
1226 if (perf_counter_overflow(counter, 1, &data)) 1229 if (perf_counter_overflow(counter, 1, &data))
1227 intel_pmu_disable_counter(&counter->hw, bit); 1230 intel_pmu_disable_counter(&counter->hw, bit);
1228 } 1231 }
@@ -1459,18 +1462,16 @@ static int intel_pmu_init(void)
1459 1462
1460static int amd_pmu_init(void) 1463static int amd_pmu_init(void)
1461{ 1464{
1465 /* Performance-monitoring supported from K7 and later: */
1466 if (boot_cpu_data.x86 < 6)
1467 return -ENODEV;
1468
1462 x86_pmu = amd_pmu; 1469 x86_pmu = amd_pmu;
1463 1470
1464 switch (boot_cpu_data.x86) { 1471 /* Events are common for all AMDs */
1465 case 0x0f: 1472 memcpy(hw_cache_event_ids, amd_hw_cache_event_ids,
1466 case 0x10: 1473 sizeof(hw_cache_event_ids));
1467 case 0x11:
1468 memcpy(hw_cache_event_ids, amd_0f_hw_cache_event_ids,
1469 sizeof(hw_cache_event_ids));
1470 1474
1471 pr_cont("AMD Family 0f/10/11 events, ");
1472 break;
1473 }
1474 return 0; 1475 return 0;
1475} 1476}
1476 1477
@@ -1554,9 +1555,9 @@ const struct pmu *hw_perf_counter_init(struct perf_counter *counter)
1554 */ 1555 */
1555 1556
1556static inline 1557static inline
1557void callchain_store(struct perf_callchain_entry *entry, unsigned long ip) 1558void callchain_store(struct perf_callchain_entry *entry, u64 ip)
1558{ 1559{
1559 if (entry->nr < MAX_STACK_DEPTH) 1560 if (entry->nr < PERF_MAX_STACK_DEPTH)
1560 entry->ip[entry->nr++] = ip; 1561 entry->ip[entry->nr++] = ip;
1561} 1562}
1562 1563
@@ -1577,8 +1578,8 @@ static void backtrace_warning(void *data, char *msg)
1577 1578
1578static int backtrace_stack(void *data, char *name) 1579static int backtrace_stack(void *data, char *name)
1579{ 1580{
1580 /* Don't bother with IRQ stacks for now */ 1581 /* Process all stacks: */
1581 return -1; 1582 return 0;
1582} 1583}
1583 1584
1584static void backtrace_address(void *data, unsigned long addr, int reliable) 1585static void backtrace_address(void *data, unsigned long addr, int reliable)
@@ -1596,47 +1597,59 @@ static const struct stacktrace_ops backtrace_ops = {
1596 .address = backtrace_address, 1597 .address = backtrace_address,
1597}; 1598};
1598 1599
1600#include "../dumpstack.h"
1601
1599static void 1602static void
1600perf_callchain_kernel(struct pt_regs *regs, struct perf_callchain_entry *entry) 1603perf_callchain_kernel(struct pt_regs *regs, struct perf_callchain_entry *entry)
1601{ 1604{
1602 unsigned long bp; 1605 callchain_store(entry, PERF_CONTEXT_KERNEL);
1603 char *stack; 1606 callchain_store(entry, regs->ip);
1604 int nr = entry->nr;
1605 1607
1606 callchain_store(entry, instruction_pointer(regs)); 1608 dump_trace(NULL, regs, NULL, 0, &backtrace_ops, entry);
1609}
1607 1610
1608 stack = ((char *)regs + sizeof(struct pt_regs)); 1611/*
1609#ifdef CONFIG_FRAME_POINTER 1612 * best effort, GUP based copy_from_user() that assumes IRQ or NMI context
1610 bp = frame_pointer(regs); 1613 */
1611#else 1614static unsigned long
1612 bp = 0; 1615copy_from_user_nmi(void *to, const void __user *from, unsigned long n)
1613#endif 1616{
1617 unsigned long offset, addr = (unsigned long)from;
1618 int type = in_nmi() ? KM_NMI : KM_IRQ0;
1619 unsigned long size, len = 0;
1620 struct page *page;
1621 void *map;
1622 int ret;
1614 1623
1615 dump_trace(NULL, regs, (void *)stack, bp, &backtrace_ops, entry); 1624 do {
1625 ret = __get_user_pages_fast(addr, 1, 0, &page);
1626 if (!ret)
1627 break;
1616 1628
1617 entry->kernel = entry->nr - nr; 1629 offset = addr & (PAGE_SIZE - 1);
1618} 1630 size = min(PAGE_SIZE - offset, n - len);
1619 1631
1632 map = kmap_atomic(page, type);
1633 memcpy(to, map+offset, size);
1634 kunmap_atomic(map, type);
1635 put_page(page);
1620 1636
1621struct stack_frame { 1637 len += size;
1622 const void __user *next_fp; 1638 to += size;
1623 unsigned long return_address; 1639 addr += size;
1624}; 1640
1641 } while (len < n);
1642
1643 return len;
1644}
1625 1645
1626static int copy_stack_frame(const void __user *fp, struct stack_frame *frame) 1646static int copy_stack_frame(const void __user *fp, struct stack_frame *frame)
1627{ 1647{
1628 int ret; 1648 unsigned long bytes;
1629 1649
1630 if (!access_ok(VERIFY_READ, fp, sizeof(*frame))) 1650 bytes = copy_from_user_nmi(frame, fp, sizeof(*frame));
1631 return 0;
1632 1651
1633 ret = 1; 1652 return bytes == sizeof(*frame);
1634 pagefault_disable();
1635 if (__copy_from_user_inatomic(frame, fp, sizeof(*frame)))
1636 ret = 0;
1637 pagefault_enable();
1638
1639 return ret;
1640} 1653}
1641 1654
1642static void 1655static void
@@ -1644,28 +1657,28 @@ perf_callchain_user(struct pt_regs *regs, struct perf_callchain_entry *entry)
1644{ 1657{
1645 struct stack_frame frame; 1658 struct stack_frame frame;
1646 const void __user *fp; 1659 const void __user *fp;
1647 int nr = entry->nr;
1648 1660
1649 regs = (struct pt_regs *)current->thread.sp0 - 1; 1661 if (!user_mode(regs))
1650 fp = (void __user *)regs->bp; 1662 regs = task_pt_regs(current);
1651 1663
1664 fp = (void __user *)regs->bp;
1665
1666 callchain_store(entry, PERF_CONTEXT_USER);
1652 callchain_store(entry, regs->ip); 1667 callchain_store(entry, regs->ip);
1653 1668
1654 while (entry->nr < MAX_STACK_DEPTH) { 1669 while (entry->nr < PERF_MAX_STACK_DEPTH) {
1655 frame.next_fp = NULL; 1670 frame.next_frame = NULL;
1656 frame.return_address = 0; 1671 frame.return_address = 0;
1657 1672
1658 if (!copy_stack_frame(fp, &frame)) 1673 if (!copy_stack_frame(fp, &frame))
1659 break; 1674 break;
1660 1675
1661 if ((unsigned long)fp < user_stack_pointer(regs)) 1676 if ((unsigned long)fp < regs->sp)
1662 break; 1677 break;
1663 1678
1664 callchain_store(entry, frame.return_address); 1679 callchain_store(entry, frame.return_address);
1665 fp = frame.next_fp; 1680 fp = frame.next_frame;
1666 } 1681 }
1667
1668 entry->user = entry->nr - nr;
1669} 1682}
1670 1683
1671static void 1684static void
@@ -1701,9 +1714,6 @@ struct perf_callchain_entry *perf_callchain(struct pt_regs *regs)
1701 entry = &__get_cpu_var(irq_entry); 1714 entry = &__get_cpu_var(irq_entry);
1702 1715
1703 entry->nr = 0; 1716 entry->nr = 0;
1704 entry->hv = 0;
1705 entry->kernel = 0;
1706 entry->user = 0;
1707 1717
1708 perf_do_callchain(regs, entry); 1718 perf_do_callchain(regs, entry);
1709 1719
diff --git a/arch/x86/mm/gup.c b/arch/x86/mm/gup.c
index f97480941269..71da1bca13cb 100644
--- a/arch/x86/mm/gup.c
+++ b/arch/x86/mm/gup.c
@@ -14,7 +14,7 @@
14static inline pte_t gup_get_pte(pte_t *ptep) 14static inline pte_t gup_get_pte(pte_t *ptep)
15{ 15{
16#ifndef CONFIG_X86_PAE 16#ifndef CONFIG_X86_PAE
17 return *ptep; 17 return ACCESS_ONCE(*ptep);
18#else 18#else
19 /* 19 /*
20 * With get_user_pages_fast, we walk down the pagetables without taking 20 * With get_user_pages_fast, we walk down the pagetables without taking
@@ -219,6 +219,62 @@ static int gup_pud_range(pgd_t pgd, unsigned long addr, unsigned long end,
219 return 1; 219 return 1;
220} 220}
221 221
222/*
223 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
224 * back to the regular GUP.
225 */
226int __get_user_pages_fast(unsigned long start, int nr_pages, int write,
227 struct page **pages)
228{
229 struct mm_struct *mm = current->mm;
230 unsigned long addr, len, end;
231 unsigned long next;
232 unsigned long flags;
233 pgd_t *pgdp;
234 int nr = 0;
235
236 start &= PAGE_MASK;
237 addr = start;
238 len = (unsigned long) nr_pages << PAGE_SHIFT;
239 end = start + len;
240 if (unlikely(!access_ok(write ? VERIFY_WRITE : VERIFY_READ,
241 (void __user *)start, len)))
242 return 0;
243
244 /*
245 * XXX: batch / limit 'nr', to avoid large irq off latency
246 * needs some instrumenting to determine the common sizes used by
247 * important workloads (eg. DB2), and whether limiting the batch size
248 * will decrease performance.
249 *
250 * It seems like we're in the clear for the moment. Direct-IO is
251 * the main guy that batches up lots of get_user_pages, and even
252 * they are limited to 64-at-a-time which is not so many.
253 */
254 /*
255 * This doesn't prevent pagetable teardown, but does prevent
256 * the pagetables and pages from being freed on x86.
257 *
258 * So long as we atomically load page table pointers versus teardown
259 * (which we do on x86, with the above PAE exception), we can follow the
260 * address down to the the page and take a ref on it.
261 */
262 local_irq_save(flags);
263 pgdp = pgd_offset(mm, addr);
264 do {
265 pgd_t pgd = *pgdp;
266
267 next = pgd_addr_end(addr, end);
268 if (pgd_none(pgd))
269 break;
270 if (!gup_pud_range(pgd, addr, next, write, pages, &nr))
271 break;
272 } while (pgdp++, addr = next, addr != end);
273 local_irq_restore(flags);
274
275 return nr;
276}
277
222/** 278/**
223 * get_user_pages_fast() - pin user pages in memory 279 * get_user_pages_fast() - pin user pages in memory
224 * @start: starting user address 280 * @start: starting user address
diff --git a/fs/anon_inodes.c b/fs/anon_inodes.c
index 1dd96d4406c0..47d4a01c5393 100644
--- a/fs/anon_inodes.c
+++ b/fs/anon_inodes.c
@@ -52,6 +52,19 @@ static const struct dentry_operations anon_inodefs_dentry_operations = {
52 .d_delete = anon_inodefs_delete_dentry, 52 .d_delete = anon_inodefs_delete_dentry,
53}; 53};
54 54
55/*
56 * nop .set_page_dirty method so that people can use .page_mkwrite on
57 * anon inodes.
58 */
59static int anon_set_page_dirty(struct page *page)
60{
61 return 0;
62};
63
64static const struct address_space_operations anon_aops = {
65 .set_page_dirty = anon_set_page_dirty,
66};
67
55/** 68/**
56 * anon_inode_getfd - creates a new file instance by hooking it up to an 69 * anon_inode_getfd - creates a new file instance by hooking it up to an
57 * anonymous inode, and a dentry that describe the "class" 70 * anonymous inode, and a dentry that describe the "class"
@@ -151,6 +164,8 @@ static struct inode *anon_inode_mkinode(void)
151 164
152 inode->i_fop = &anon_inode_fops; 165 inode->i_fop = &anon_inode_fops;
153 166
167 inode->i_mapping->a_ops = &anon_aops;
168
154 /* 169 /*
155 * Mark the inode dirty from the very beginning, 170 * Mark the inode dirty from the very beginning,
156 * that way it will never be moved to the dirty 171 * that way it will never be moved to the dirty
diff --git a/include/asm-generic/kmap_types.h b/include/asm-generic/kmap_types.h
index 54e8b3d956b7..eddbce0f9fb9 100644
--- a/include/asm-generic/kmap_types.h
+++ b/include/asm-generic/kmap_types.h
@@ -24,7 +24,10 @@ D(12) KM_SOFTIRQ1,
24D(13) KM_SYNC_ICACHE, 24D(13) KM_SYNC_ICACHE,
25D(14) KM_SYNC_DCACHE, 25D(14) KM_SYNC_DCACHE,
26D(15) KM_UML_USERCOPY, /* UML specific, for copy_*_user - used in do_op_one_page */ 26D(15) KM_UML_USERCOPY, /* UML specific, for copy_*_user - used in do_op_one_page */
27D(16) KM_TYPE_NR 27D(16) KM_IRQ_PTE,
28D(17) KM_NMI,
29D(18) KM_NMI_PTE,
30D(19) KM_TYPE_NR
28}; 31};
29 32
30#undef D 33#undef D
diff --git a/include/linux/mm.h b/include/linux/mm.h
index d88d6fc530ad..cf260d848eb9 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -854,6 +854,12 @@ extern int mprotect_fixup(struct vm_area_struct *vma,
854 unsigned long end, unsigned long newflags); 854 unsigned long end, unsigned long newflags);
855 855
856/* 856/*
857 * doesn't attempt to fault and will return short.
858 */
859int __get_user_pages_fast(unsigned long start, int nr_pages, int write,
860 struct page **pages);
861
862/*
857 * A callback you can register to apply pressure to ageable caches. 863 * A callback you can register to apply pressure to ageable caches.
858 * 864 *
859 * 'shrink' is passed a count 'nr_to_scan' and a 'gfpmask'. It should 865 * 'shrink' is passed a count 'nr_to_scan' and a 'gfpmask'. It should
diff --git a/include/linux/perf_counter.h b/include/linux/perf_counter.h
index 1b3118a1023a..89698d8aba5c 100644
--- a/include/linux/perf_counter.h
+++ b/include/linux/perf_counter.h
@@ -236,10 +236,16 @@ struct perf_counter_mmap_page {
236 /* 236 /*
237 * Control data for the mmap() data buffer. 237 * Control data for the mmap() data buffer.
238 * 238 *
239 * User-space reading this value should issue an rmb(), on SMP capable 239 * User-space reading the @data_head value should issue an rmb(), on
240 * platforms, after reading this value -- see perf_counter_wakeup(). 240 * SMP capable platforms, after reading this value -- see
241 * perf_counter_wakeup().
242 *
243 * When the mapping is PROT_WRITE the @data_tail value should be
244 * written by userspace to reflect the last read data. In this case
245 * the kernel will not over-write unread data.
241 */ 246 */
242 __u64 data_head; /* head in the data section */ 247 __u64 data_head; /* head in the data section */
248 __u64 data_tail; /* user-space written tail */
243}; 249};
244 250
245#define PERF_EVENT_MISC_CPUMODE_MASK (3 << 0) 251#define PERF_EVENT_MISC_CPUMODE_MASK (3 << 0)
@@ -275,6 +281,15 @@ enum perf_event_type {
275 281
276 /* 282 /*
277 * struct { 283 * struct {
284 * struct perf_event_header header;
285 * u64 id;
286 * u64 lost;
287 * };
288 */
289 PERF_EVENT_LOST = 2,
290
291 /*
292 * struct {
278 * struct perf_event_header header; 293 * struct perf_event_header header;
279 * 294 *
280 * u32 pid, tid; 295 * u32 pid, tid;
@@ -313,30 +328,39 @@ enum perf_event_type {
313 328
314 /* 329 /*
315 * When header.misc & PERF_EVENT_MISC_OVERFLOW the event_type field 330 * When header.misc & PERF_EVENT_MISC_OVERFLOW the event_type field
316 * will be PERF_RECORD_* 331 * will be PERF_SAMPLE_*
317 * 332 *
318 * struct { 333 * struct {
319 * struct perf_event_header header; 334 * struct perf_event_header header;
320 * 335 *
321 * { u64 ip; } && PERF_RECORD_IP 336 * { u64 ip; } && PERF_SAMPLE_IP
322 * { u32 pid, tid; } && PERF_RECORD_TID 337 * { u32 pid, tid; } && PERF_SAMPLE_TID
323 * { u64 time; } && PERF_RECORD_TIME 338 * { u64 time; } && PERF_SAMPLE_TIME
324 * { u64 addr; } && PERF_RECORD_ADDR 339 * { u64 addr; } && PERF_SAMPLE_ADDR
325 * { u64 config; } && PERF_RECORD_CONFIG 340 * { u64 config; } && PERF_SAMPLE_CONFIG
326 * { u32 cpu, res; } && PERF_RECORD_CPU 341 * { u32 cpu, res; } && PERF_SAMPLE_CPU
327 * 342 *
328 * { u64 nr; 343 * { u64 nr;
329 * { u64 id, val; } cnt[nr]; } && PERF_RECORD_GROUP 344 * { u64 id, val; } cnt[nr]; } && PERF_SAMPLE_GROUP
330 * 345 *
331 * { u16 nr, 346 * { u64 nr,
332 * hv, 347 * u64 ips[nr]; } && PERF_SAMPLE_CALLCHAIN
333 * kernel,
334 * user;
335 * u64 ips[nr]; } && PERF_RECORD_CALLCHAIN
336 * }; 348 * };
337 */ 349 */
338}; 350};
339 351
352enum perf_callchain_context {
353 PERF_CONTEXT_HV = (__u64)-32,
354 PERF_CONTEXT_KERNEL = (__u64)-128,
355 PERF_CONTEXT_USER = (__u64)-512,
356
357 PERF_CONTEXT_GUEST = (__u64)-2048,
358 PERF_CONTEXT_GUEST_KERNEL = (__u64)-2176,
359 PERF_CONTEXT_GUEST_USER = (__u64)-2560,
360
361 PERF_CONTEXT_MAX = (__u64)-4095,
362};
363
340#ifdef __KERNEL__ 364#ifdef __KERNEL__
341/* 365/*
342 * Kernel-internal data types and definitions: 366 * Kernel-internal data types and definitions:
@@ -356,6 +380,13 @@ enum perf_event_type {
356#include <linux/pid_namespace.h> 380#include <linux/pid_namespace.h>
357#include <asm/atomic.h> 381#include <asm/atomic.h>
358 382
383#define PERF_MAX_STACK_DEPTH 255
384
385struct perf_callchain_entry {
386 __u64 nr;
387 __u64 ip[PERF_MAX_STACK_DEPTH];
388};
389
359struct task_struct; 390struct task_struct;
360 391
361/** 392/**
@@ -414,6 +445,7 @@ struct file;
414struct perf_mmap_data { 445struct perf_mmap_data {
415 struct rcu_head rcu_head; 446 struct rcu_head rcu_head;
416 int nr_pages; /* nr of data pages */ 447 int nr_pages; /* nr of data pages */
448 int writable; /* are we writable */
417 int nr_locked; /* nr pages mlocked */ 449 int nr_locked; /* nr pages mlocked */
418 450
419 atomic_t poll; /* POLL_ for wakeups */ 451 atomic_t poll; /* POLL_ for wakeups */
@@ -423,8 +455,8 @@ struct perf_mmap_data {
423 atomic_long_t done_head; /* completed head */ 455 atomic_long_t done_head; /* completed head */
424 456
425 atomic_t lock; /* concurrent writes */ 457 atomic_t lock; /* concurrent writes */
426
427 atomic_t wakeup; /* needs a wakeup */ 458 atomic_t wakeup; /* needs a wakeup */
459 atomic_t lost; /* nr records lost */
428 460
429 struct perf_counter_mmap_page *user_page; 461 struct perf_counter_mmap_page *user_page;
430 void *data_pages[0]; 462 void *data_pages[0];
@@ -604,6 +636,7 @@ extern void perf_counter_task_tick(struct task_struct *task, int cpu);
604extern int perf_counter_init_task(struct task_struct *child); 636extern int perf_counter_init_task(struct task_struct *child);
605extern void perf_counter_exit_task(struct task_struct *child); 637extern void perf_counter_exit_task(struct task_struct *child);
606extern void perf_counter_free_task(struct task_struct *task); 638extern void perf_counter_free_task(struct task_struct *task);
639extern void set_perf_counter_pending(void);
607extern void perf_counter_do_pending(void); 640extern void perf_counter_do_pending(void);
608extern void perf_counter_print_debug(void); 641extern void perf_counter_print_debug(void);
609extern void __perf_disable(void); 642extern void __perf_disable(void);
@@ -649,18 +682,6 @@ static inline void perf_counter_mmap(struct vm_area_struct *vma)
649extern void perf_counter_comm(struct task_struct *tsk); 682extern void perf_counter_comm(struct task_struct *tsk);
650extern void perf_counter_fork(struct task_struct *tsk); 683extern void perf_counter_fork(struct task_struct *tsk);
651 684
652extern void perf_counter_task_migration(struct task_struct *task, int cpu);
653
654#define MAX_STACK_DEPTH 255
655
656struct perf_callchain_entry {
657 u16 nr;
658 u16 hv;
659 u16 kernel;
660 u16 user;
661 u64 ip[MAX_STACK_DEPTH];
662};
663
664extern struct perf_callchain_entry *perf_callchain(struct pt_regs *regs); 685extern struct perf_callchain_entry *perf_callchain(struct pt_regs *regs);
665 686
666extern int sysctl_perf_counter_paranoid; 687extern int sysctl_perf_counter_paranoid;
@@ -701,8 +722,6 @@ static inline void perf_counter_mmap(struct vm_area_struct *vma) { }
701static inline void perf_counter_comm(struct task_struct *tsk) { } 722static inline void perf_counter_comm(struct task_struct *tsk) { }
702static inline void perf_counter_fork(struct task_struct *tsk) { } 723static inline void perf_counter_fork(struct task_struct *tsk) { }
703static inline void perf_counter_init(void) { } 724static inline void perf_counter_init(void) { }
704static inline void perf_counter_task_migration(struct task_struct *task,
705 int cpu) { }
706#endif 725#endif
707 726
708#endif /* __KERNEL__ */ 727#endif /* __KERNEL__ */
diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c
index 29b685f551aa..1a933a221ea4 100644
--- a/kernel/perf_counter.c
+++ b/kernel/perf_counter.c
@@ -124,7 +124,7 @@ void perf_enable(void)
124 124
125static void get_ctx(struct perf_counter_context *ctx) 125static void get_ctx(struct perf_counter_context *ctx)
126{ 126{
127 atomic_inc(&ctx->refcount); 127 WARN_ON(!atomic_inc_not_zero(&ctx->refcount));
128} 128}
129 129
130static void free_ctx(struct rcu_head *head) 130static void free_ctx(struct rcu_head *head)
@@ -175,6 +175,11 @@ perf_lock_task_context(struct task_struct *task, unsigned long *flags)
175 spin_unlock_irqrestore(&ctx->lock, *flags); 175 spin_unlock_irqrestore(&ctx->lock, *flags);
176 goto retry; 176 goto retry;
177 } 177 }
178
179 if (!atomic_inc_not_zero(&ctx->refcount)) {
180 spin_unlock_irqrestore(&ctx->lock, *flags);
181 ctx = NULL;
182 }
178 } 183 }
179 rcu_read_unlock(); 184 rcu_read_unlock();
180 return ctx; 185 return ctx;
@@ -193,7 +198,6 @@ static struct perf_counter_context *perf_pin_task_context(struct task_struct *ta
193 ctx = perf_lock_task_context(task, &flags); 198 ctx = perf_lock_task_context(task, &flags);
194 if (ctx) { 199 if (ctx) {
195 ++ctx->pin_count; 200 ++ctx->pin_count;
196 get_ctx(ctx);
197 spin_unlock_irqrestore(&ctx->lock, flags); 201 spin_unlock_irqrestore(&ctx->lock, flags);
198 } 202 }
199 return ctx; 203 return ctx;
@@ -1283,7 +1287,7 @@ static void perf_ctx_adjust_freq(struct perf_counter_context *ctx)
1283 if (!interrupts) { 1287 if (!interrupts) {
1284 perf_disable(); 1288 perf_disable();
1285 counter->pmu->disable(counter); 1289 counter->pmu->disable(counter);
1286 atomic_set(&hwc->period_left, 0); 1290 atomic64_set(&hwc->period_left, 0);
1287 counter->pmu->enable(counter); 1291 counter->pmu->enable(counter);
1288 perf_enable(); 1292 perf_enable();
1289 } 1293 }
@@ -1459,11 +1463,6 @@ static struct perf_counter_context *find_get_context(pid_t pid, int cpu)
1459 put_ctx(parent_ctx); 1463 put_ctx(parent_ctx);
1460 ctx->parent_ctx = NULL; /* no longer a clone */ 1464 ctx->parent_ctx = NULL; /* no longer a clone */
1461 } 1465 }
1462 /*
1463 * Get an extra reference before dropping the lock so that
1464 * this context won't get freed if the task exits.
1465 */
1466 get_ctx(ctx);
1467 spin_unlock_irqrestore(&ctx->lock, flags); 1466 spin_unlock_irqrestore(&ctx->lock, flags);
1468 } 1467 }
1469 1468
@@ -1553,7 +1552,7 @@ static int perf_release(struct inode *inode, struct file *file)
1553static ssize_t 1552static ssize_t
1554perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count) 1553perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count)
1555{ 1554{
1556 u64 values[3]; 1555 u64 values[4];
1557 int n; 1556 int n;
1558 1557
1559 /* 1558 /*
@@ -1620,22 +1619,6 @@ static void perf_counter_reset(struct perf_counter *counter)
1620 perf_counter_update_userpage(counter); 1619 perf_counter_update_userpage(counter);
1621} 1620}
1622 1621
1623static void perf_counter_for_each_sibling(struct perf_counter *counter,
1624 void (*func)(struct perf_counter *))
1625{
1626 struct perf_counter_context *ctx = counter->ctx;
1627 struct perf_counter *sibling;
1628
1629 WARN_ON_ONCE(ctx->parent_ctx);
1630 mutex_lock(&ctx->mutex);
1631 counter = counter->group_leader;
1632
1633 func(counter);
1634 list_for_each_entry(sibling, &counter->sibling_list, list_entry)
1635 func(sibling);
1636 mutex_unlock(&ctx->mutex);
1637}
1638
1639/* 1622/*
1640 * Holding the top-level counter's child_mutex means that any 1623 * Holding the top-level counter's child_mutex means that any
1641 * descendant process that has inherited this counter will block 1624 * descendant process that has inherited this counter will block
@@ -1658,14 +1641,18 @@ static void perf_counter_for_each_child(struct perf_counter *counter,
1658static void perf_counter_for_each(struct perf_counter *counter, 1641static void perf_counter_for_each(struct perf_counter *counter,
1659 void (*func)(struct perf_counter *)) 1642 void (*func)(struct perf_counter *))
1660{ 1643{
1661 struct perf_counter *child; 1644 struct perf_counter_context *ctx = counter->ctx;
1645 struct perf_counter *sibling;
1662 1646
1663 WARN_ON_ONCE(counter->ctx->parent_ctx); 1647 WARN_ON_ONCE(ctx->parent_ctx);
1664 mutex_lock(&counter->child_mutex); 1648 mutex_lock(&ctx->mutex);
1665 perf_counter_for_each_sibling(counter, func); 1649 counter = counter->group_leader;
1666 list_for_each_entry(child, &counter->child_list, child_list) 1650
1667 perf_counter_for_each_sibling(child, func); 1651 perf_counter_for_each_child(counter, func);
1668 mutex_unlock(&counter->child_mutex); 1652 func(counter);
1653 list_for_each_entry(sibling, &counter->sibling_list, list_entry)
1654 perf_counter_for_each_child(counter, func);
1655 mutex_unlock(&ctx->mutex);
1669} 1656}
1670 1657
1671static int perf_counter_period(struct perf_counter *counter, u64 __user *arg) 1658static int perf_counter_period(struct perf_counter *counter, u64 __user *arg)
@@ -1806,6 +1793,12 @@ static int perf_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1806 struct perf_mmap_data *data; 1793 struct perf_mmap_data *data;
1807 int ret = VM_FAULT_SIGBUS; 1794 int ret = VM_FAULT_SIGBUS;
1808 1795
1796 if (vmf->flags & FAULT_FLAG_MKWRITE) {
1797 if (vmf->pgoff == 0)
1798 ret = 0;
1799 return ret;
1800 }
1801
1809 rcu_read_lock(); 1802 rcu_read_lock();
1810 data = rcu_dereference(counter->data); 1803 data = rcu_dereference(counter->data);
1811 if (!data) 1804 if (!data)
@@ -1819,9 +1812,16 @@ static int perf_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1819 if ((unsigned)nr > data->nr_pages) 1812 if ((unsigned)nr > data->nr_pages)
1820 goto unlock; 1813 goto unlock;
1821 1814
1815 if (vmf->flags & FAULT_FLAG_WRITE)
1816 goto unlock;
1817
1822 vmf->page = virt_to_page(data->data_pages[nr]); 1818 vmf->page = virt_to_page(data->data_pages[nr]);
1823 } 1819 }
1820
1824 get_page(vmf->page); 1821 get_page(vmf->page);
1822 vmf->page->mapping = vma->vm_file->f_mapping;
1823 vmf->page->index = vmf->pgoff;
1824
1825 ret = 0; 1825 ret = 0;
1826unlock: 1826unlock:
1827 rcu_read_unlock(); 1827 rcu_read_unlock();
@@ -1874,6 +1874,14 @@ fail:
1874 return -ENOMEM; 1874 return -ENOMEM;
1875} 1875}
1876 1876
1877static void perf_mmap_free_page(unsigned long addr)
1878{
1879 struct page *page = virt_to_page(addr);
1880
1881 page->mapping = NULL;
1882 __free_page(page);
1883}
1884
1877static void __perf_mmap_data_free(struct rcu_head *rcu_head) 1885static void __perf_mmap_data_free(struct rcu_head *rcu_head)
1878{ 1886{
1879 struct perf_mmap_data *data; 1887 struct perf_mmap_data *data;
@@ -1881,9 +1889,10 @@ static void __perf_mmap_data_free(struct rcu_head *rcu_head)
1881 1889
1882 data = container_of(rcu_head, struct perf_mmap_data, rcu_head); 1890 data = container_of(rcu_head, struct perf_mmap_data, rcu_head);
1883 1891
1884 free_page((unsigned long)data->user_page); 1892 perf_mmap_free_page((unsigned long)data->user_page);
1885 for (i = 0; i < data->nr_pages; i++) 1893 for (i = 0; i < data->nr_pages; i++)
1886 free_page((unsigned long)data->data_pages[i]); 1894 perf_mmap_free_page((unsigned long)data->data_pages[i]);
1895
1887 kfree(data); 1896 kfree(data);
1888} 1897}
1889 1898
@@ -1920,9 +1929,10 @@ static void perf_mmap_close(struct vm_area_struct *vma)
1920} 1929}
1921 1930
1922static struct vm_operations_struct perf_mmap_vmops = { 1931static struct vm_operations_struct perf_mmap_vmops = {
1923 .open = perf_mmap_open, 1932 .open = perf_mmap_open,
1924 .close = perf_mmap_close, 1933 .close = perf_mmap_close,
1925 .fault = perf_mmap_fault, 1934 .fault = perf_mmap_fault,
1935 .page_mkwrite = perf_mmap_fault,
1926}; 1936};
1927 1937
1928static int perf_mmap(struct file *file, struct vm_area_struct *vma) 1938static int perf_mmap(struct file *file, struct vm_area_struct *vma)
@@ -1936,7 +1946,7 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma)
1936 long user_extra, extra; 1946 long user_extra, extra;
1937 int ret = 0; 1947 int ret = 0;
1938 1948
1939 if (!(vma->vm_flags & VM_SHARED) || (vma->vm_flags & VM_WRITE)) 1949 if (!(vma->vm_flags & VM_SHARED))
1940 return -EINVAL; 1950 return -EINVAL;
1941 1951
1942 vma_size = vma->vm_end - vma->vm_start; 1952 vma_size = vma->vm_end - vma->vm_start;
@@ -1995,10 +2005,12 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma)
1995 atomic_long_add(user_extra, &user->locked_vm); 2005 atomic_long_add(user_extra, &user->locked_vm);
1996 vma->vm_mm->locked_vm += extra; 2006 vma->vm_mm->locked_vm += extra;
1997 counter->data->nr_locked = extra; 2007 counter->data->nr_locked = extra;
2008 if (vma->vm_flags & VM_WRITE)
2009 counter->data->writable = 1;
2010
1998unlock: 2011unlock:
1999 mutex_unlock(&counter->mmap_mutex); 2012 mutex_unlock(&counter->mmap_mutex);
2000 2013
2001 vma->vm_flags &= ~VM_MAYWRITE;
2002 vma->vm_flags |= VM_RESERVED; 2014 vma->vm_flags |= VM_RESERVED;
2003 vma->vm_ops = &perf_mmap_vmops; 2015 vma->vm_ops = &perf_mmap_vmops;
2004 2016
@@ -2175,11 +2187,38 @@ struct perf_output_handle {
2175 unsigned long head; 2187 unsigned long head;
2176 unsigned long offset; 2188 unsigned long offset;
2177 int nmi; 2189 int nmi;
2178 int overflow; 2190 int sample;
2179 int locked; 2191 int locked;
2180 unsigned long flags; 2192 unsigned long flags;
2181}; 2193};
2182 2194
2195static bool perf_output_space(struct perf_mmap_data *data,
2196 unsigned int offset, unsigned int head)
2197{
2198 unsigned long tail;
2199 unsigned long mask;
2200
2201 if (!data->writable)
2202 return true;
2203
2204 mask = (data->nr_pages << PAGE_SHIFT) - 1;
2205 /*
2206 * Userspace could choose to issue a mb() before updating the tail
2207 * pointer. So that all reads will be completed before the write is
2208 * issued.
2209 */
2210 tail = ACCESS_ONCE(data->user_page->data_tail);
2211 smp_rmb();
2212
2213 offset = (offset - tail) & mask;
2214 head = (head - tail) & mask;
2215
2216 if ((int)(head - offset) < 0)
2217 return false;
2218
2219 return true;
2220}
2221
2183static void perf_output_wakeup(struct perf_output_handle *handle) 2222static void perf_output_wakeup(struct perf_output_handle *handle)
2184{ 2223{
2185 atomic_set(&handle->data->poll, POLL_IN); 2224 atomic_set(&handle->data->poll, POLL_IN);
@@ -2270,12 +2309,57 @@ out:
2270 local_irq_restore(handle->flags); 2309 local_irq_restore(handle->flags);
2271} 2310}
2272 2311
2312static void perf_output_copy(struct perf_output_handle *handle,
2313 const void *buf, unsigned int len)
2314{
2315 unsigned int pages_mask;
2316 unsigned int offset;
2317 unsigned int size;
2318 void **pages;
2319
2320 offset = handle->offset;
2321 pages_mask = handle->data->nr_pages - 1;
2322 pages = handle->data->data_pages;
2323
2324 do {
2325 unsigned int page_offset;
2326 int nr;
2327
2328 nr = (offset >> PAGE_SHIFT) & pages_mask;
2329 page_offset = offset & (PAGE_SIZE - 1);
2330 size = min_t(unsigned int, PAGE_SIZE - page_offset, len);
2331
2332 memcpy(pages[nr] + page_offset, buf, size);
2333
2334 len -= size;
2335 buf += size;
2336 offset += size;
2337 } while (len);
2338
2339 handle->offset = offset;
2340
2341 /*
2342 * Check we didn't copy past our reservation window, taking the
2343 * possible unsigned int wrap into account.
2344 */
2345 WARN_ON_ONCE(((long)(handle->head - handle->offset)) < 0);
2346}
2347
2348#define perf_output_put(handle, x) \
2349 perf_output_copy((handle), &(x), sizeof(x))
2350
2273static int perf_output_begin(struct perf_output_handle *handle, 2351static int perf_output_begin(struct perf_output_handle *handle,
2274 struct perf_counter *counter, unsigned int size, 2352 struct perf_counter *counter, unsigned int size,
2275 int nmi, int overflow) 2353 int nmi, int sample)
2276{ 2354{
2277 struct perf_mmap_data *data; 2355 struct perf_mmap_data *data;
2278 unsigned int offset, head; 2356 unsigned int offset, head;
2357 int have_lost;
2358 struct {
2359 struct perf_event_header header;
2360 u64 id;
2361 u64 lost;
2362 } lost_event;
2279 2363
2280 /* 2364 /*
2281 * For inherited counters we send all the output towards the parent. 2365 * For inherited counters we send all the output towards the parent.
@@ -2288,19 +2372,25 @@ static int perf_output_begin(struct perf_output_handle *handle,
2288 if (!data) 2372 if (!data)
2289 goto out; 2373 goto out;
2290 2374
2291 handle->data = data; 2375 handle->data = data;
2292 handle->counter = counter; 2376 handle->counter = counter;
2293 handle->nmi = nmi; 2377 handle->nmi = nmi;
2294 handle->overflow = overflow; 2378 handle->sample = sample;
2295 2379
2296 if (!data->nr_pages) 2380 if (!data->nr_pages)
2297 goto fail; 2381 goto fail;
2298 2382
2383 have_lost = atomic_read(&data->lost);
2384 if (have_lost)
2385 size += sizeof(lost_event);
2386
2299 perf_output_lock(handle); 2387 perf_output_lock(handle);
2300 2388
2301 do { 2389 do {
2302 offset = head = atomic_long_read(&data->head); 2390 offset = head = atomic_long_read(&data->head);
2303 head += size; 2391 head += size;
2392 if (unlikely(!perf_output_space(data, offset, head)))
2393 goto fail;
2304 } while (atomic_long_cmpxchg(&data->head, offset, head) != offset); 2394 } while (atomic_long_cmpxchg(&data->head, offset, head) != offset);
2305 2395
2306 handle->offset = offset; 2396 handle->offset = offset;
@@ -2309,55 +2399,27 @@ static int perf_output_begin(struct perf_output_handle *handle,
2309 if ((offset >> PAGE_SHIFT) != (head >> PAGE_SHIFT)) 2399 if ((offset >> PAGE_SHIFT) != (head >> PAGE_SHIFT))
2310 atomic_set(&data->wakeup, 1); 2400 atomic_set(&data->wakeup, 1);
2311 2401
2402 if (have_lost) {
2403 lost_event.header.type = PERF_EVENT_LOST;
2404 lost_event.header.misc = 0;
2405 lost_event.header.size = sizeof(lost_event);
2406 lost_event.id = counter->id;
2407 lost_event.lost = atomic_xchg(&data->lost, 0);
2408
2409 perf_output_put(handle, lost_event);
2410 }
2411
2312 return 0; 2412 return 0;
2313 2413
2314fail: 2414fail:
2315 perf_output_wakeup(handle); 2415 atomic_inc(&data->lost);
2416 perf_output_unlock(handle);
2316out: 2417out:
2317 rcu_read_unlock(); 2418 rcu_read_unlock();
2318 2419
2319 return -ENOSPC; 2420 return -ENOSPC;
2320} 2421}
2321 2422
2322static void perf_output_copy(struct perf_output_handle *handle,
2323 const void *buf, unsigned int len)
2324{
2325 unsigned int pages_mask;
2326 unsigned int offset;
2327 unsigned int size;
2328 void **pages;
2329
2330 offset = handle->offset;
2331 pages_mask = handle->data->nr_pages - 1;
2332 pages = handle->data->data_pages;
2333
2334 do {
2335 unsigned int page_offset;
2336 int nr;
2337
2338 nr = (offset >> PAGE_SHIFT) & pages_mask;
2339 page_offset = offset & (PAGE_SIZE - 1);
2340 size = min_t(unsigned int, PAGE_SIZE - page_offset, len);
2341
2342 memcpy(pages[nr] + page_offset, buf, size);
2343
2344 len -= size;
2345 buf += size;
2346 offset += size;
2347 } while (len);
2348
2349 handle->offset = offset;
2350
2351 /*
2352 * Check we didn't copy past our reservation window, taking the
2353 * possible unsigned int wrap into account.
2354 */
2355 WARN_ON_ONCE(((long)(handle->head - handle->offset)) < 0);
2356}
2357
2358#define perf_output_put(handle, x) \
2359 perf_output_copy((handle), &(x), sizeof(x))
2360
2361static void perf_output_end(struct perf_output_handle *handle) 2423static void perf_output_end(struct perf_output_handle *handle)
2362{ 2424{
2363 struct perf_counter *counter = handle->counter; 2425 struct perf_counter *counter = handle->counter;
@@ -2365,7 +2427,7 @@ static void perf_output_end(struct perf_output_handle *handle)
2365 2427
2366 int wakeup_events = counter->attr.wakeup_events; 2428 int wakeup_events = counter->attr.wakeup_events;
2367 2429
2368 if (handle->overflow && wakeup_events) { 2430 if (handle->sample && wakeup_events) {
2369 int events = atomic_inc_return(&data->events); 2431 int events = atomic_inc_return(&data->events);
2370 if (events >= wakeup_events) { 2432 if (events >= wakeup_events) {
2371 atomic_sub(wakeup_events, &data->events); 2433 atomic_sub(wakeup_events, &data->events);
@@ -2970,7 +3032,7 @@ static void perf_log_throttle(struct perf_counter *counter, int enable)
2970} 3032}
2971 3033
2972/* 3034/*
2973 * Generic counter overflow handling. 3035 * Generic counter overflow handling, sampling.
2974 */ 3036 */
2975 3037
2976int perf_counter_overflow(struct perf_counter *counter, int nmi, 3038int perf_counter_overflow(struct perf_counter *counter, int nmi,
@@ -3109,20 +3171,15 @@ static enum hrtimer_restart perf_swcounter_hrtimer(struct hrtimer *hrtimer)
3109} 3171}
3110 3172
3111static void perf_swcounter_overflow(struct perf_counter *counter, 3173static void perf_swcounter_overflow(struct perf_counter *counter,
3112 int nmi, struct pt_regs *regs, u64 addr) 3174 int nmi, struct perf_sample_data *data)
3113{ 3175{
3114 struct perf_sample_data data = { 3176 data->period = counter->hw.last_period;
3115 .regs = regs,
3116 .addr = addr,
3117 .period = counter->hw.last_period,
3118 };
3119 3177
3120 perf_swcounter_update(counter); 3178 perf_swcounter_update(counter);
3121 perf_swcounter_set_period(counter); 3179 perf_swcounter_set_period(counter);
3122 if (perf_counter_overflow(counter, nmi, &data)) 3180 if (perf_counter_overflow(counter, nmi, data))
3123 /* soft-disable the counter */ 3181 /* soft-disable the counter */
3124 ; 3182 ;
3125
3126} 3183}
3127 3184
3128static int perf_swcounter_is_counting(struct perf_counter *counter) 3185static int perf_swcounter_is_counting(struct perf_counter *counter)
@@ -3187,18 +3244,18 @@ static int perf_swcounter_match(struct perf_counter *counter,
3187} 3244}
3188 3245
3189static void perf_swcounter_add(struct perf_counter *counter, u64 nr, 3246static void perf_swcounter_add(struct perf_counter *counter, u64 nr,
3190 int nmi, struct pt_regs *regs, u64 addr) 3247 int nmi, struct perf_sample_data *data)
3191{ 3248{
3192 int neg = atomic64_add_negative(nr, &counter->hw.count); 3249 int neg = atomic64_add_negative(nr, &counter->hw.count);
3193 3250
3194 if (counter->hw.sample_period && !neg && regs) 3251 if (counter->hw.sample_period && !neg && data->regs)
3195 perf_swcounter_overflow(counter, nmi, regs, addr); 3252 perf_swcounter_overflow(counter, nmi, data);
3196} 3253}
3197 3254
3198static void perf_swcounter_ctx_event(struct perf_counter_context *ctx, 3255static void perf_swcounter_ctx_event(struct perf_counter_context *ctx,
3199 enum perf_type_id type, u32 event, 3256 enum perf_type_id type,
3200 u64 nr, int nmi, struct pt_regs *regs, 3257 u32 event, u64 nr, int nmi,
3201 u64 addr) 3258 struct perf_sample_data *data)
3202{ 3259{
3203 struct perf_counter *counter; 3260 struct perf_counter *counter;
3204 3261
@@ -3207,8 +3264,8 @@ static void perf_swcounter_ctx_event(struct perf_counter_context *ctx,
3207 3264
3208 rcu_read_lock(); 3265 rcu_read_lock();
3209 list_for_each_entry_rcu(counter, &ctx->event_list, event_entry) { 3266 list_for_each_entry_rcu(counter, &ctx->event_list, event_entry) {
3210 if (perf_swcounter_match(counter, type, event, regs)) 3267 if (perf_swcounter_match(counter, type, event, data->regs))
3211 perf_swcounter_add(counter, nr, nmi, regs, addr); 3268 perf_swcounter_add(counter, nr, nmi, data);
3212 } 3269 }
3213 rcu_read_unlock(); 3270 rcu_read_unlock();
3214} 3271}
@@ -3227,9 +3284,9 @@ static int *perf_swcounter_recursion_context(struct perf_cpu_context *cpuctx)
3227 return &cpuctx->recursion[0]; 3284 return &cpuctx->recursion[0];
3228} 3285}
3229 3286
3230static void __perf_swcounter_event(enum perf_type_id type, u32 event, 3287static void do_perf_swcounter_event(enum perf_type_id type, u32 event,
3231 u64 nr, int nmi, struct pt_regs *regs, 3288 u64 nr, int nmi,
3232 u64 addr) 3289 struct perf_sample_data *data)
3233{ 3290{
3234 struct perf_cpu_context *cpuctx = &get_cpu_var(perf_cpu_context); 3291 struct perf_cpu_context *cpuctx = &get_cpu_var(perf_cpu_context);
3235 int *recursion = perf_swcounter_recursion_context(cpuctx); 3292 int *recursion = perf_swcounter_recursion_context(cpuctx);
@@ -3242,7 +3299,7 @@ static void __perf_swcounter_event(enum perf_type_id type, u32 event,
3242 barrier(); 3299 barrier();
3243 3300
3244 perf_swcounter_ctx_event(&cpuctx->ctx, type, event, 3301 perf_swcounter_ctx_event(&cpuctx->ctx, type, event,
3245 nr, nmi, regs, addr); 3302 nr, nmi, data);
3246 rcu_read_lock(); 3303 rcu_read_lock();
3247 /* 3304 /*
3248 * doesn't really matter which of the child contexts the 3305 * doesn't really matter which of the child contexts the
@@ -3250,7 +3307,7 @@ static void __perf_swcounter_event(enum perf_type_id type, u32 event,
3250 */ 3307 */
3251 ctx = rcu_dereference(current->perf_counter_ctxp); 3308 ctx = rcu_dereference(current->perf_counter_ctxp);
3252 if (ctx) 3309 if (ctx)
3253 perf_swcounter_ctx_event(ctx, type, event, nr, nmi, regs, addr); 3310 perf_swcounter_ctx_event(ctx, type, event, nr, nmi, data);
3254 rcu_read_unlock(); 3311 rcu_read_unlock();
3255 3312
3256 barrier(); 3313 barrier();
@@ -3263,7 +3320,12 @@ out:
3263void 3320void
3264perf_swcounter_event(u32 event, u64 nr, int nmi, struct pt_regs *regs, u64 addr) 3321perf_swcounter_event(u32 event, u64 nr, int nmi, struct pt_regs *regs, u64 addr)
3265{ 3322{
3266 __perf_swcounter_event(PERF_TYPE_SOFTWARE, event, nr, nmi, regs, addr); 3323 struct perf_sample_data data = {
3324 .regs = regs,
3325 .addr = addr,
3326 };
3327
3328 do_perf_swcounter_event(PERF_TYPE_SOFTWARE, event, nr, nmi, &data);
3267} 3329}
3268 3330
3269static void perf_swcounter_read(struct perf_counter *counter) 3331static void perf_swcounter_read(struct perf_counter *counter)
@@ -3404,36 +3466,18 @@ static const struct pmu perf_ops_task_clock = {
3404 .read = task_clock_perf_counter_read, 3466 .read = task_clock_perf_counter_read,
3405}; 3467};
3406 3468
3407/*
3408 * Software counter: cpu migrations
3409 */
3410void perf_counter_task_migration(struct task_struct *task, int cpu)
3411{
3412 struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu);
3413 struct perf_counter_context *ctx;
3414
3415 perf_swcounter_ctx_event(&cpuctx->ctx, PERF_TYPE_SOFTWARE,
3416 PERF_COUNT_SW_CPU_MIGRATIONS,
3417 1, 1, NULL, 0);
3418
3419 ctx = perf_pin_task_context(task);
3420 if (ctx) {
3421 perf_swcounter_ctx_event(ctx, PERF_TYPE_SOFTWARE,
3422 PERF_COUNT_SW_CPU_MIGRATIONS,
3423 1, 1, NULL, 0);
3424 perf_unpin_context(ctx);
3425 }
3426}
3427
3428#ifdef CONFIG_EVENT_PROFILE 3469#ifdef CONFIG_EVENT_PROFILE
3429void perf_tpcounter_event(int event_id) 3470void perf_tpcounter_event(int event_id)
3430{ 3471{
3431 struct pt_regs *regs = get_irq_regs(); 3472 struct perf_sample_data data = {
3473 .regs = get_irq_regs();
3474 .addr = 0,
3475 };
3432 3476
3433 if (!regs) 3477 if (!data.regs)
3434 regs = task_pt_regs(current); 3478 data.regs = task_pt_regs(current);
3435 3479
3436 __perf_swcounter_event(PERF_TYPE_TRACEPOINT, event_id, 1, 1, regs, 0); 3480 do_perf_swcounter_event(PERF_TYPE_TRACEPOINT, event_id, 1, 1, &data);
3437} 3481}
3438EXPORT_SYMBOL_GPL(perf_tpcounter_event); 3482EXPORT_SYMBOL_GPL(perf_tpcounter_event);
3439 3483
diff --git a/kernel/sched.c b/kernel/sched.c
index 92e51287b980..7c9098d186e6 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -1978,7 +1978,8 @@ void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
1978 if (task_hot(p, old_rq->clock, NULL)) 1978 if (task_hot(p, old_rq->clock, NULL))
1979 schedstat_inc(p, se.nr_forced2_migrations); 1979 schedstat_inc(p, se.nr_forced2_migrations);
1980#endif 1980#endif
1981 perf_counter_task_migration(p, new_cpu); 1981 perf_swcounter_event(PERF_COUNT_SW_CPU_MIGRATIONS,
1982 1, 1, NULL, 0);
1982 } 1983 }
1983 p->se.vruntime -= old_cfsrq->min_vruntime - 1984 p->se.vruntime -= old_cfsrq->min_vruntime -
1984 new_cfsrq->min_vruntime; 1985 new_cfsrq->min_vruntime;
diff --git a/tools/perf/Makefile b/tools/perf/Makefile
index 0cbd5d6874ec..36d7eef49913 100644
--- a/tools/perf/Makefile
+++ b/tools/perf/Makefile
@@ -157,10 +157,15 @@ uname_R := $(shell sh -c 'uname -r 2>/dev/null || echo not')
157uname_P := $(shell sh -c 'uname -p 2>/dev/null || echo not') 157uname_P := $(shell sh -c 'uname -p 2>/dev/null || echo not')
158uname_V := $(shell sh -c 'uname -v 2>/dev/null || echo not') 158uname_V := $(shell sh -c 'uname -v 2>/dev/null || echo not')
159 159
160# If we're on a 64-bit kernel, use -m64
161ifneq ($(patsubst %64,%,$(uname_M)),$(uname_M))
162 M64 := -m64
163endif
164
160# CFLAGS and LDFLAGS are for the users to override from the command line. 165# CFLAGS and LDFLAGS are for the users to override from the command line.
161 166
162CFLAGS = -ggdb3 -Wall -Werror -Wstrict-prototypes -Wmissing-declarations -Wmissing-prototypes -std=gnu99 -Wdeclaration-after-statement -O6 167CFLAGS = $(M64) -ggdb3 -Wall -Wstrict-prototypes -Wmissing-declarations -Wmissing-prototypes -std=gnu99 -Wdeclaration-after-statement -Werror -O6
163LDFLAGS = -lpthread -lrt -lelf 168LDFLAGS = -lpthread -lrt -lelf -lm
164ALL_CFLAGS = $(CFLAGS) 169ALL_CFLAGS = $(CFLAGS)
165ALL_LDFLAGS = $(LDFLAGS) 170ALL_LDFLAGS = $(LDFLAGS)
166STRIP ?= strip 171STRIP ?= strip
@@ -285,6 +290,7 @@ LIB_FILE=libperf.a
285 290
286LIB_H += ../../include/linux/perf_counter.h 291LIB_H += ../../include/linux/perf_counter.h
287LIB_H += perf.h 292LIB_H += perf.h
293LIB_H += types.h
288LIB_H += util/list.h 294LIB_H += util/list.h
289LIB_H += util/rbtree.h 295LIB_H += util/rbtree.h
290LIB_H += util/levenshtein.h 296LIB_H += util/levenshtein.h
diff --git a/tools/perf/builtin-annotate.c b/tools/perf/builtin-annotate.c
index b1ed5f766cb3..7e58e3ad1508 100644
--- a/tools/perf/builtin-annotate.c
+++ b/tools/perf/builtin-annotate.c
@@ -25,6 +25,10 @@
25#define SHOW_USER 2 25#define SHOW_USER 2
26#define SHOW_HV 4 26#define SHOW_HV 4
27 27
28#define MIN_GREEN 0.5
29#define MIN_RED 5.0
30
31
28static char const *input_name = "perf.data"; 32static char const *input_name = "perf.data";
29static char *vmlinux = "vmlinux"; 33static char *vmlinux = "vmlinux";
30 34
@@ -39,40 +43,42 @@ static int dump_trace = 0;
39 43
40static int verbose; 44static int verbose;
41 45
46static int print_line;
47
42static unsigned long page_size; 48static unsigned long page_size;
43static unsigned long mmap_window = 32; 49static unsigned long mmap_window = 32;
44 50
45struct ip_event { 51struct ip_event {
46 struct perf_event_header header; 52 struct perf_event_header header;
47 __u64 ip; 53 u64 ip;
48 __u32 pid, tid; 54 u32 pid, tid;
49}; 55};
50 56
51struct mmap_event { 57struct mmap_event {
52 struct perf_event_header header; 58 struct perf_event_header header;
53 __u32 pid, tid; 59 u32 pid, tid;
54 __u64 start; 60 u64 start;
55 __u64 len; 61 u64 len;
56 __u64 pgoff; 62 u64 pgoff;
57 char filename[PATH_MAX]; 63 char filename[PATH_MAX];
58}; 64};
59 65
60struct comm_event { 66struct comm_event {
61 struct perf_event_header header; 67 struct perf_event_header header;
62 __u32 pid, tid; 68 u32 pid, tid;
63 char comm[16]; 69 char comm[16];
64}; 70};
65 71
66struct fork_event { 72struct fork_event {
67 struct perf_event_header header; 73 struct perf_event_header header;
68 __u32 pid, ppid; 74 u32 pid, ppid;
69}; 75};
70 76
71struct period_event { 77struct period_event {
72 struct perf_event_header header; 78 struct perf_event_header header;
73 __u64 time; 79 u64 time;
74 __u64 id; 80 u64 id;
75 __u64 sample_period; 81 u64 sample_period;
76}; 82};
77 83
78typedef union event_union { 84typedef union event_union {
@@ -84,6 +90,13 @@ typedef union event_union {
84 struct period_event period; 90 struct period_event period;
85} event_t; 91} event_t;
86 92
93
94struct sym_ext {
95 struct rb_node node;
96 double percent;
97 char *path;
98};
99
87static LIST_HEAD(dsos); 100static LIST_HEAD(dsos);
88static struct dso *kernel_dso; 101static struct dso *kernel_dso;
89static struct dso *vdso; 102static struct dso *vdso;
@@ -145,7 +158,7 @@ static void dsos__fprintf(FILE *fp)
145 dso__fprintf(pos, fp); 158 dso__fprintf(pos, fp);
146} 159}
147 160
148static struct symbol *vdso__find_symbol(struct dso *dso, __u64 ip) 161static struct symbol *vdso__find_symbol(struct dso *dso, u64 ip)
149{ 162{
150 return dso__find_symbol(kernel_dso, ip); 163 return dso__find_symbol(kernel_dso, ip);
151} 164}
@@ -178,19 +191,19 @@ static int load_kernel(void)
178 191
179struct map { 192struct map {
180 struct list_head node; 193 struct list_head node;
181 __u64 start; 194 u64 start;
182 __u64 end; 195 u64 end;
183 __u64 pgoff; 196 u64 pgoff;
184 __u64 (*map_ip)(struct map *, __u64); 197 u64 (*map_ip)(struct map *, u64);
185 struct dso *dso; 198 struct dso *dso;
186}; 199};
187 200
188static __u64 map__map_ip(struct map *map, __u64 ip) 201static u64 map__map_ip(struct map *map, u64 ip)
189{ 202{
190 return ip - map->start + map->pgoff; 203 return ip - map->start + map->pgoff;
191} 204}
192 205
193static __u64 vdso__map_ip(struct map *map, __u64 ip) 206static u64 vdso__map_ip(struct map *map, u64 ip)
194{ 207{
195 return ip; 208 return ip;
196} 209}
@@ -373,7 +386,7 @@ static int thread__fork(struct thread *self, struct thread *parent)
373 return 0; 386 return 0;
374} 387}
375 388
376static struct map *thread__find_map(struct thread *self, __u64 ip) 389static struct map *thread__find_map(struct thread *self, u64 ip)
377{ 390{
378 struct map *pos; 391 struct map *pos;
379 392
@@ -414,7 +427,7 @@ struct hist_entry {
414 struct map *map; 427 struct map *map;
415 struct dso *dso; 428 struct dso *dso;
416 struct symbol *sym; 429 struct symbol *sym;
417 __u64 ip; 430 u64 ip;
418 char level; 431 char level;
419 432
420 uint32_t count; 433 uint32_t count;
@@ -519,7 +532,7 @@ sort__dso_print(FILE *fp, struct hist_entry *self)
519 if (self->dso) 532 if (self->dso)
520 return fprintf(fp, "%-25s", self->dso->name); 533 return fprintf(fp, "%-25s", self->dso->name);
521 534
522 return fprintf(fp, "%016llx ", (__u64)self->ip); 535 return fprintf(fp, "%016llx ", (u64)self->ip);
523} 536}
524 537
525static struct sort_entry sort_dso = { 538static struct sort_entry sort_dso = {
@@ -533,7 +546,7 @@ static struct sort_entry sort_dso = {
533static int64_t 546static int64_t
534sort__sym_cmp(struct hist_entry *left, struct hist_entry *right) 547sort__sym_cmp(struct hist_entry *left, struct hist_entry *right)
535{ 548{
536 __u64 ip_l, ip_r; 549 u64 ip_l, ip_r;
537 550
538 if (left->sym == right->sym) 551 if (left->sym == right->sym)
539 return 0; 552 return 0;
@@ -550,13 +563,13 @@ sort__sym_print(FILE *fp, struct hist_entry *self)
550 size_t ret = 0; 563 size_t ret = 0;
551 564
552 if (verbose) 565 if (verbose)
553 ret += fprintf(fp, "%#018llx ", (__u64)self->ip); 566 ret += fprintf(fp, "%#018llx ", (u64)self->ip);
554 567
555 if (self->sym) { 568 if (self->sym) {
556 ret += fprintf(fp, "[%c] %s", 569 ret += fprintf(fp, "[%c] %s",
557 self->dso == kernel_dso ? 'k' : '.', self->sym->name); 570 self->dso == kernel_dso ? 'k' : '.', self->sym->name);
558 } else { 571 } else {
559 ret += fprintf(fp, "%#016llx", (__u64)self->ip); 572 ret += fprintf(fp, "%#016llx", (u64)self->ip);
560 } 573 }
561 574
562 return ret; 575 return ret;
@@ -647,7 +660,7 @@ hist_entry__collapse(struct hist_entry *left, struct hist_entry *right)
647/* 660/*
648 * collect histogram counts 661 * collect histogram counts
649 */ 662 */
650static void hist_hit(struct hist_entry *he, __u64 ip) 663static void hist_hit(struct hist_entry *he, u64 ip)
651{ 664{
652 unsigned int sym_size, offset; 665 unsigned int sym_size, offset;
653 struct symbol *sym = he->sym; 666 struct symbol *sym = he->sym;
@@ -676,7 +689,7 @@ static void hist_hit(struct hist_entry *he, __u64 ip)
676 689
677static int 690static int
678hist_entry__add(struct thread *thread, struct map *map, struct dso *dso, 691hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
679 struct symbol *sym, __u64 ip, char level) 692 struct symbol *sym, u64 ip, char level)
680{ 693{
681 struct rb_node **p = &hist.rb_node; 694 struct rb_node **p = &hist.rb_node;
682 struct rb_node *parent = NULL; 695 struct rb_node *parent = NULL;
@@ -848,7 +861,7 @@ process_overflow_event(event_t *event, unsigned long offset, unsigned long head)
848 int show = 0; 861 int show = 0;
849 struct dso *dso = NULL; 862 struct dso *dso = NULL;
850 struct thread *thread = threads__findnew(event->ip.pid); 863 struct thread *thread = threads__findnew(event->ip.pid);
851 __u64 ip = event->ip.ip; 864 u64 ip = event->ip.ip;
852 struct map *map = NULL; 865 struct map *map = NULL;
853 866
854 dprintf("%p [%p]: PERF_EVENT (IP, %d): %d: %p\n", 867 dprintf("%p [%p]: PERF_EVENT (IP, %d): %d: %p\n",
@@ -1030,13 +1043,33 @@ process_event(event_t *event, unsigned long offset, unsigned long head)
1030 return 0; 1043 return 0;
1031} 1044}
1032 1045
1046static char *get_color(double percent)
1047{
1048 char *color = PERF_COLOR_NORMAL;
1049
1050 /*
1051 * We color high-overhead entries in red, mid-overhead
1052 * entries in green - and keep the low overhead places
1053 * normal:
1054 */
1055 if (percent >= MIN_RED)
1056 color = PERF_COLOR_RED;
1057 else {
1058 if (percent > MIN_GREEN)
1059 color = PERF_COLOR_GREEN;
1060 }
1061 return color;
1062}
1063
1033static int 1064static int
1034parse_line(FILE *file, struct symbol *sym, __u64 start, __u64 len) 1065parse_line(FILE *file, struct symbol *sym, u64 start, u64 len)
1035{ 1066{
1036 char *line = NULL, *tmp, *tmp2; 1067 char *line = NULL, *tmp, *tmp2;
1068 static const char *prev_line;
1069 static const char *prev_color;
1037 unsigned int offset; 1070 unsigned int offset;
1038 size_t line_len; 1071 size_t line_len;
1039 __u64 line_ip; 1072 u64 line_ip;
1040 int ret; 1073 int ret;
1041 char *c; 1074 char *c;
1042 1075
@@ -1073,27 +1106,36 @@ parse_line(FILE *file, struct symbol *sym, __u64 start, __u64 len)
1073 } 1106 }
1074 1107
1075 if (line_ip != -1) { 1108 if (line_ip != -1) {
1109 const char *path = NULL;
1076 unsigned int hits = 0; 1110 unsigned int hits = 0;
1077 double percent = 0.0; 1111 double percent = 0.0;
1078 char *color = PERF_COLOR_NORMAL; 1112 char *color;
1113 struct sym_ext *sym_ext = sym->priv;
1079 1114
1080 offset = line_ip - start; 1115 offset = line_ip - start;
1081 if (offset < len) 1116 if (offset < len)
1082 hits = sym->hist[offset]; 1117 hits = sym->hist[offset];
1083 1118
1084 if (sym->hist_sum) 1119 if (offset < len && sym_ext) {
1120 path = sym_ext[offset].path;
1121 percent = sym_ext[offset].percent;
1122 } else if (sym->hist_sum)
1085 percent = 100.0 * hits / sym->hist_sum; 1123 percent = 100.0 * hits / sym->hist_sum;
1086 1124
1125 color = get_color(percent);
1126
1087 /* 1127 /*
1088 * We color high-overhead entries in red, mid-overhead 1128 * Also color the filename and line if needed, with
1089 * entries in green - and keep the low overhead places 1129 * the same color than the percentage. Don't print it
1090 * normal: 1130 * twice for close colored ip with the same filename:line
1091 */ 1131 */
1092 if (percent >= 5.0) 1132 if (path) {
1093 color = PERF_COLOR_RED; 1133 if (!prev_line || strcmp(prev_line, path)
1094 else { 1134 || color != prev_color) {
1095 if (percent > 0.5) 1135 color_fprintf(stdout, color, " %s", path);
1096 color = PERF_COLOR_GREEN; 1136 prev_line = path;
1137 prev_color = color;
1138 }
1097 } 1139 }
1098 1140
1099 color_fprintf(stdout, color, " %7.2f", percent); 1141 color_fprintf(stdout, color, " %7.2f", percent);
@@ -1109,10 +1151,125 @@ parse_line(FILE *file, struct symbol *sym, __u64 start, __u64 len)
1109 return 0; 1151 return 0;
1110} 1152}
1111 1153
1154static struct rb_root root_sym_ext;
1155
1156static void insert_source_line(struct sym_ext *sym_ext)
1157{
1158 struct sym_ext *iter;
1159 struct rb_node **p = &root_sym_ext.rb_node;
1160 struct rb_node *parent = NULL;
1161
1162 while (*p != NULL) {
1163 parent = *p;
1164 iter = rb_entry(parent, struct sym_ext, node);
1165
1166 if (sym_ext->percent > iter->percent)
1167 p = &(*p)->rb_left;
1168 else
1169 p = &(*p)->rb_right;
1170 }
1171
1172 rb_link_node(&sym_ext->node, parent, p);
1173 rb_insert_color(&sym_ext->node, &root_sym_ext);
1174}
1175
1176static void free_source_line(struct symbol *sym, int len)
1177{
1178 struct sym_ext *sym_ext = sym->priv;
1179 int i;
1180
1181 if (!sym_ext)
1182 return;
1183
1184 for (i = 0; i < len; i++)
1185 free(sym_ext[i].path);
1186 free(sym_ext);
1187
1188 sym->priv = NULL;
1189 root_sym_ext = RB_ROOT;
1190}
1191
1192/* Get the filename:line for the colored entries */
1193static void
1194get_source_line(struct symbol *sym, u64 start, int len, char *filename)
1195{
1196 int i;
1197 char cmd[PATH_MAX * 2];
1198 struct sym_ext *sym_ext;
1199
1200 if (!sym->hist_sum)
1201 return;
1202
1203 sym->priv = calloc(len, sizeof(struct sym_ext));
1204 if (!sym->priv)
1205 return;
1206
1207 sym_ext = sym->priv;
1208
1209 for (i = 0; i < len; i++) {
1210 char *path = NULL;
1211 size_t line_len;
1212 u64 offset;
1213 FILE *fp;
1214
1215 sym_ext[i].percent = 100.0 * sym->hist[i] / sym->hist_sum;
1216 if (sym_ext[i].percent <= 0.5)
1217 continue;
1218
1219 offset = start + i;
1220 sprintf(cmd, "addr2line -e %s %016llx", filename, offset);
1221 fp = popen(cmd, "r");
1222 if (!fp)
1223 continue;
1224
1225 if (getline(&path, &line_len, fp) < 0 || !line_len)
1226 goto next;
1227
1228 sym_ext[i].path = malloc(sizeof(char) * line_len + 1);
1229 if (!sym_ext[i].path)
1230 goto next;
1231
1232 strcpy(sym_ext[i].path, path);
1233 insert_source_line(&sym_ext[i]);
1234
1235 next:
1236 pclose(fp);
1237 }
1238}
1239
1240static void print_summary(char *filename)
1241{
1242 struct sym_ext *sym_ext;
1243 struct rb_node *node;
1244
1245 printf("\nSorted summary for file %s\n", filename);
1246 printf("----------------------------------------------\n\n");
1247
1248 if (RB_EMPTY_ROOT(&root_sym_ext)) {
1249 printf(" Nothing higher than %1.1f%%\n", MIN_GREEN);
1250 return;
1251 }
1252
1253 node = rb_first(&root_sym_ext);
1254 while (node) {
1255 double percent;
1256 char *color;
1257 char *path;
1258
1259 sym_ext = rb_entry(node, struct sym_ext, node);
1260 percent = sym_ext->percent;
1261 color = get_color(percent);
1262 path = sym_ext->path;
1263
1264 color_fprintf(stdout, color, " %7.2f %s", percent, path);
1265 node = rb_next(node);
1266 }
1267}
1268
1112static void annotate_sym(struct dso *dso, struct symbol *sym) 1269static void annotate_sym(struct dso *dso, struct symbol *sym)
1113{ 1270{
1114 char *filename = dso->name; 1271 char *filename = dso->name;
1115 __u64 start, end, len; 1272 u64 start, end, len;
1116 char command[PATH_MAX*2]; 1273 char command[PATH_MAX*2];
1117 FILE *file; 1274 FILE *file;
1118 1275
@@ -1121,13 +1278,6 @@ static void annotate_sym(struct dso *dso, struct symbol *sym)
1121 if (dso == kernel_dso) 1278 if (dso == kernel_dso)
1122 filename = vmlinux; 1279 filename = vmlinux;
1123 1280
1124 printf("\n------------------------------------------------\n");
1125 printf(" Percent | Source code & Disassembly of %s\n", filename);
1126 printf("------------------------------------------------\n");
1127
1128 if (verbose >= 2)
1129 printf("annotating [%p] %30s : [%p] %30s\n", dso, dso->name, sym, sym->name);
1130
1131 start = sym->obj_start; 1281 start = sym->obj_start;
1132 if (!start) 1282 if (!start)
1133 start = sym->start; 1283 start = sym->start;
@@ -1135,7 +1285,19 @@ static void annotate_sym(struct dso *dso, struct symbol *sym)
1135 end = start + sym->end - sym->start + 1; 1285 end = start + sym->end - sym->start + 1;
1136 len = sym->end - sym->start; 1286 len = sym->end - sym->start;
1137 1287
1138 sprintf(command, "objdump --start-address=0x%016Lx --stop-address=0x%016Lx -dS %s", (__u64)start, (__u64)end, filename); 1288 if (print_line) {
1289 get_source_line(sym, start, len, filename);
1290 print_summary(filename);
1291 }
1292
1293 printf("\n\n------------------------------------------------\n");
1294 printf(" Percent | Source code & Disassembly of %s\n", filename);
1295 printf("------------------------------------------------\n");
1296
1297 if (verbose >= 2)
1298 printf("annotating [%p] %30s : [%p] %30s\n", dso, dso->name, sym, sym->name);
1299
1300 sprintf(command, "objdump --start-address=0x%016Lx --stop-address=0x%016Lx -dS %s", (u64)start, (u64)end, filename);
1139 1301
1140 if (verbose >= 3) 1302 if (verbose >= 3)
1141 printf("doing: %s\n", command); 1303 printf("doing: %s\n", command);
@@ -1150,6 +1312,8 @@ static void annotate_sym(struct dso *dso, struct symbol *sym)
1150 } 1312 }
1151 1313
1152 pclose(file); 1314 pclose(file);
1315 if (print_line)
1316 free_source_line(sym, len);
1153} 1317}
1154 1318
1155static void find_annotations(void) 1319static void find_annotations(void)
@@ -1308,6 +1472,8 @@ static const struct option options[] = {
1308 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 1472 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1309 "dump raw trace in ASCII"), 1473 "dump raw trace in ASCII"),
1310 OPT_STRING('k', "vmlinux", &vmlinux, "file", "vmlinux pathname"), 1474 OPT_STRING('k', "vmlinux", &vmlinux, "file", "vmlinux pathname"),
1475 OPT_BOOLEAN('l', "print-line", &print_line,
1476 "print matching source lines (may be slow)"),
1311 OPT_END() 1477 OPT_END()
1312}; 1478};
1313 1479
diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 0f5771f615da..d7ebbd757543 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -37,33 +37,37 @@ static pid_t target_pid = -1;
37static int inherit = 1; 37static int inherit = 1;
38static int force = 0; 38static int force = 0;
39static int append_file = 0; 39static int append_file = 0;
40static int call_graph = 0;
40static int verbose = 0; 41static int verbose = 0;
41 42
42static long samples; 43static long samples;
43static struct timeval last_read; 44static struct timeval last_read;
44static struct timeval this_read; 45static struct timeval this_read;
45 46
46static __u64 bytes_written; 47static u64 bytes_written;
47 48
48static struct pollfd event_array[MAX_NR_CPUS * MAX_COUNTERS]; 49static struct pollfd event_array[MAX_NR_CPUS * MAX_COUNTERS];
49 50
50static int nr_poll; 51static int nr_poll;
51static int nr_cpu; 52static int nr_cpu;
52 53
54static int file_new = 1;
55static struct perf_file_header file_header;
56
53struct mmap_event { 57struct mmap_event {
54 struct perf_event_header header; 58 struct perf_event_header header;
55 __u32 pid; 59 u32 pid;
56 __u32 tid; 60 u32 tid;
57 __u64 start; 61 u64 start;
58 __u64 len; 62 u64 len;
59 __u64 pgoff; 63 u64 pgoff;
60 char filename[PATH_MAX]; 64 char filename[PATH_MAX];
61}; 65};
62 66
63struct comm_event { 67struct comm_event {
64 struct perf_event_header header; 68 struct perf_event_header header;
65 __u32 pid; 69 u32 pid;
66 __u32 tid; 70 u32 tid;
67 char comm[16]; 71 char comm[16];
68}; 72};
69 73
@@ -77,10 +81,10 @@ struct mmap_data {
77 81
78static struct mmap_data mmap_array[MAX_NR_CPUS][MAX_COUNTERS]; 82static struct mmap_data mmap_array[MAX_NR_CPUS][MAX_COUNTERS];
79 83
80static unsigned int mmap_read_head(struct mmap_data *md) 84static unsigned long mmap_read_head(struct mmap_data *md)
81{ 85{
82 struct perf_counter_mmap_page *pc = md->base; 86 struct perf_counter_mmap_page *pc = md->base;
83 int head; 87 long head;
84 88
85 head = pc->data_head; 89 head = pc->data_head;
86 rmb(); 90 rmb();
@@ -88,6 +92,32 @@ static unsigned int mmap_read_head(struct mmap_data *md)
88 return head; 92 return head;
89} 93}
90 94
95static void mmap_write_tail(struct mmap_data *md, unsigned long tail)
96{
97 struct perf_counter_mmap_page *pc = md->base;
98
99 /*
100 * ensure all reads are done before we write the tail out.
101 */
102 /* mb(); */
103 pc->data_tail = tail;
104}
105
106static void write_output(void *buf, size_t size)
107{
108 while (size) {
109 int ret = write(output, buf, size);
110
111 if (ret < 0)
112 die("failed to write");
113
114 size -= ret;
115 buf += ret;
116
117 bytes_written += ret;
118 }
119}
120
91static void mmap_read(struct mmap_data *md) 121static void mmap_read(struct mmap_data *md)
92{ 122{
93 unsigned int head = mmap_read_head(md); 123 unsigned int head = mmap_read_head(md);
@@ -108,7 +138,7 @@ static void mmap_read(struct mmap_data *md)
108 * In either case, truncate and restart at head. 138 * In either case, truncate and restart at head.
109 */ 139 */
110 diff = head - old; 140 diff = head - old;
111 if (diff > md->mask / 2 || diff < 0) { 141 if (diff < 0) {
112 struct timeval iv; 142 struct timeval iv;
113 unsigned long msecs; 143 unsigned long msecs;
114 144
@@ -136,36 +166,17 @@ static void mmap_read(struct mmap_data *md)
136 size = md->mask + 1 - (old & md->mask); 166 size = md->mask + 1 - (old & md->mask);
137 old += size; 167 old += size;
138 168
139 while (size) { 169 write_output(buf, size);
140 int ret = write(output, buf, size);
141
142 if (ret < 0)
143 die("failed to write");
144
145 size -= ret;
146 buf += ret;
147
148 bytes_written += ret;
149 }
150 } 170 }
151 171
152 buf = &data[old & md->mask]; 172 buf = &data[old & md->mask];
153 size = head - old; 173 size = head - old;
154 old += size; 174 old += size;
155 175
156 while (size) { 176 write_output(buf, size);
157 int ret = write(output, buf, size);
158
159 if (ret < 0)
160 die("failed to write");
161
162 size -= ret;
163 buf += ret;
164
165 bytes_written += ret;
166 }
167 177
168 md->prev = old; 178 md->prev = old;
179 mmap_write_tail(md, old);
169} 180}
170 181
171static volatile int done = 0; 182static volatile int done = 0;
@@ -191,7 +202,7 @@ static void pid_synthesize_comm_event(pid_t pid, int full)
191 struct comm_event comm_ev; 202 struct comm_event comm_ev;
192 char filename[PATH_MAX]; 203 char filename[PATH_MAX];
193 char bf[BUFSIZ]; 204 char bf[BUFSIZ];
194 int fd, ret; 205 int fd;
195 size_t size; 206 size_t size;
196 char *field, *sep; 207 char *field, *sep;
197 DIR *tasks; 208 DIR *tasks;
@@ -201,8 +212,12 @@ static void pid_synthesize_comm_event(pid_t pid, int full)
201 212
202 fd = open(filename, O_RDONLY); 213 fd = open(filename, O_RDONLY);
203 if (fd < 0) { 214 if (fd < 0) {
204 fprintf(stderr, "couldn't open %s\n", filename); 215 /*
205 exit(EXIT_FAILURE); 216 * We raced with a task exiting - just return:
217 */
218 if (verbose)
219 fprintf(stderr, "couldn't open %s\n", filename);
220 return;
206 } 221 }
207 if (read(fd, bf, sizeof(bf)) < 0) { 222 if (read(fd, bf, sizeof(bf)) < 0) {
208 fprintf(stderr, "couldn't read %s\n", filename); 223 fprintf(stderr, "couldn't read %s\n", filename);
@@ -223,17 +238,13 @@ static void pid_synthesize_comm_event(pid_t pid, int full)
223 238
224 comm_ev.pid = pid; 239 comm_ev.pid = pid;
225 comm_ev.header.type = PERF_EVENT_COMM; 240 comm_ev.header.type = PERF_EVENT_COMM;
226 size = ALIGN(size, sizeof(__u64)); 241 size = ALIGN(size, sizeof(u64));
227 comm_ev.header.size = sizeof(comm_ev) - (sizeof(comm_ev.comm) - size); 242 comm_ev.header.size = sizeof(comm_ev) - (sizeof(comm_ev.comm) - size);
228 243
229 if (!full) { 244 if (!full) {
230 comm_ev.tid = pid; 245 comm_ev.tid = pid;
231 246
232 ret = write(output, &comm_ev, comm_ev.header.size); 247 write_output(&comm_ev, comm_ev.header.size);
233 if (ret < 0) {
234 perror("failed to write");
235 exit(-1);
236 }
237 return; 248 return;
238 } 249 }
239 250
@@ -248,11 +259,7 @@ static void pid_synthesize_comm_event(pid_t pid, int full)
248 259
249 comm_ev.tid = pid; 260 comm_ev.tid = pid;
250 261
251 ret = write(output, &comm_ev, comm_ev.header.size); 262 write_output(&comm_ev, comm_ev.header.size);
252 if (ret < 0) {
253 perror("failed to write");
254 exit(-1);
255 }
256 } 263 }
257 closedir(tasks); 264 closedir(tasks);
258 return; 265 return;
@@ -272,8 +279,12 @@ static void pid_synthesize_mmap_samples(pid_t pid)
272 279
273 fp = fopen(filename, "r"); 280 fp = fopen(filename, "r");
274 if (fp == NULL) { 281 if (fp == NULL) {
275 fprintf(stderr, "couldn't open %s\n", filename); 282 /*
276 exit(EXIT_FAILURE); 283 * We raced with a task exiting - just return:
284 */
285 if (verbose)
286 fprintf(stderr, "couldn't open %s\n", filename);
287 return;
277 } 288 }
278 while (1) { 289 while (1) {
279 char bf[BUFSIZ], *pbf = bf; 290 char bf[BUFSIZ], *pbf = bf;
@@ -304,17 +315,14 @@ static void pid_synthesize_mmap_samples(pid_t pid)
304 size = strlen(execname); 315 size = strlen(execname);
305 execname[size - 1] = '\0'; /* Remove \n */ 316 execname[size - 1] = '\0'; /* Remove \n */
306 memcpy(mmap_ev.filename, execname, size); 317 memcpy(mmap_ev.filename, execname, size);
307 size = ALIGN(size, sizeof(__u64)); 318 size = ALIGN(size, sizeof(u64));
308 mmap_ev.len -= mmap_ev.start; 319 mmap_ev.len -= mmap_ev.start;
309 mmap_ev.header.size = (sizeof(mmap_ev) - 320 mmap_ev.header.size = (sizeof(mmap_ev) -
310 (sizeof(mmap_ev.filename) - size)); 321 (sizeof(mmap_ev.filename) - size));
311 mmap_ev.pid = pid; 322 mmap_ev.pid = pid;
312 mmap_ev.tid = pid; 323 mmap_ev.tid = pid;
313 324
314 if (write(output, &mmap_ev, mmap_ev.header.size) < 0) { 325 write_output(&mmap_ev, mmap_ev.header.size);
315 perror("failed to write");
316 exit(-1);
317 }
318 } 326 }
319 } 327 }
320 328
@@ -351,11 +359,25 @@ static void create_counter(int counter, int cpu, pid_t pid)
351 int track = 1; 359 int track = 1;
352 360
353 attr->sample_type = PERF_SAMPLE_IP | PERF_SAMPLE_TID; 361 attr->sample_type = PERF_SAMPLE_IP | PERF_SAMPLE_TID;
362
354 if (freq) { 363 if (freq) {
355 attr->sample_type |= PERF_SAMPLE_PERIOD; 364 attr->sample_type |= PERF_SAMPLE_PERIOD;
356 attr->freq = 1; 365 attr->freq = 1;
357 attr->sample_freq = freq; 366 attr->sample_freq = freq;
358 } 367 }
368
369 if (call_graph)
370 attr->sample_type |= PERF_SAMPLE_CALLCHAIN;
371
372 if (file_new) {
373 file_header.sample_type = attr->sample_type;
374 } else {
375 if (file_header.sample_type != attr->sample_type) {
376 fprintf(stderr, "incompatible append\n");
377 exit(-1);
378 }
379 }
380
359 attr->mmap = track; 381 attr->mmap = track;
360 attr->comm = track; 382 attr->comm = track;
361 attr->inherit = (cpu < 0) && inherit; 383 attr->inherit = (cpu < 0) && inherit;
@@ -410,7 +432,7 @@ try_again:
410 mmap_array[nr_cpu][counter].prev = 0; 432 mmap_array[nr_cpu][counter].prev = 0;
411 mmap_array[nr_cpu][counter].mask = mmap_pages*page_size - 1; 433 mmap_array[nr_cpu][counter].mask = mmap_pages*page_size - 1;
412 mmap_array[nr_cpu][counter].base = mmap(NULL, (mmap_pages+1)*page_size, 434 mmap_array[nr_cpu][counter].base = mmap(NULL, (mmap_pages+1)*page_size,
413 PROT_READ, MAP_SHARED, fd[nr_cpu][counter], 0); 435 PROT_READ|PROT_WRITE, MAP_SHARED, fd[nr_cpu][counter], 0);
414 if (mmap_array[nr_cpu][counter].base == MAP_FAILED) { 436 if (mmap_array[nr_cpu][counter].base == MAP_FAILED) {
415 error("failed to mmap with %d (%s)\n", errno, strerror(errno)); 437 error("failed to mmap with %d (%s)\n", errno, strerror(errno));
416 exit(-1); 438 exit(-1);
@@ -435,6 +457,14 @@ static void open_counters(int cpu, pid_t pid)
435 nr_cpu++; 457 nr_cpu++;
436} 458}
437 459
460static void atexit_header(void)
461{
462 file_header.data_size += bytes_written;
463
464 if (pwrite(output, &file_header, sizeof(file_header), 0) == -1)
465 perror("failed to write on file headers");
466}
467
438static int __cmd_record(int argc, const char **argv) 468static int __cmd_record(int argc, const char **argv)
439{ 469{
440 int i, counter; 470 int i, counter;
@@ -448,6 +478,10 @@ static int __cmd_record(int argc, const char **argv)
448 assert(nr_cpus <= MAX_NR_CPUS); 478 assert(nr_cpus <= MAX_NR_CPUS);
449 assert(nr_cpus >= 0); 479 assert(nr_cpus >= 0);
450 480
481 atexit(sig_atexit);
482 signal(SIGCHLD, sig_handler);
483 signal(SIGINT, sig_handler);
484
451 if (!stat(output_name, &st) && !force && !append_file) { 485 if (!stat(output_name, &st) && !force && !append_file) {
452 fprintf(stderr, "Error, output file %s exists, use -A to append or -f to overwrite.\n", 486 fprintf(stderr, "Error, output file %s exists, use -A to append or -f to overwrite.\n",
453 output_name); 487 output_name);
@@ -456,7 +490,7 @@ static int __cmd_record(int argc, const char **argv)
456 490
457 flags = O_CREAT|O_RDWR; 491 flags = O_CREAT|O_RDWR;
458 if (append_file) 492 if (append_file)
459 flags |= O_APPEND; 493 file_new = 0;
460 else 494 else
461 flags |= O_TRUNC; 495 flags |= O_TRUNC;
462 496
@@ -466,15 +500,22 @@ static int __cmd_record(int argc, const char **argv)
466 exit(-1); 500 exit(-1);
467 } 501 }
468 502
503 if (!file_new) {
504 if (read(output, &file_header, sizeof(file_header)) == -1) {
505 perror("failed to read file headers");
506 exit(-1);
507 }
508
509 lseek(output, file_header.data_size, SEEK_CUR);
510 }
511
512 atexit(atexit_header);
513
469 if (!system_wide) { 514 if (!system_wide) {
470 open_counters(-1, target_pid != -1 ? target_pid : getpid()); 515 open_counters(-1, target_pid != -1 ? target_pid : getpid());
471 } else for (i = 0; i < nr_cpus; i++) 516 } else for (i = 0; i < nr_cpus; i++)
472 open_counters(i, target_pid); 517 open_counters(i, target_pid);
473 518
474 atexit(sig_atexit);
475 signal(SIGCHLD, sig_handler);
476 signal(SIGINT, sig_handler);
477
478 if (target_pid == -1 && argc) { 519 if (target_pid == -1 && argc) {
479 pid = fork(); 520 pid = fork();
480 if (pid < 0) 521 if (pid < 0)
@@ -555,6 +596,8 @@ static const struct option options[] = {
555 "profile at this frequency"), 596 "profile at this frequency"),
556 OPT_INTEGER('m', "mmap-pages", &mmap_pages, 597 OPT_INTEGER('m', "mmap-pages", &mmap_pages,
557 "number of mmap data pages"), 598 "number of mmap data pages"),
599 OPT_BOOLEAN('g', "call-graph", &call_graph,
600 "do call-graph (stack chain/backtrace) recording"),
558 OPT_BOOLEAN('v', "verbose", &verbose, 601 OPT_BOOLEAN('v', "verbose", &verbose,
559 "be more verbose (show counter open errors, etc)"), 602 "be more verbose (show counter open errors, etc)"),
560 OPT_END() 603 OPT_END()
diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c
index 82fa93b4db99..5eb5566f0c95 100644
--- a/tools/perf/builtin-report.c
+++ b/tools/perf/builtin-report.c
@@ -36,45 +36,65 @@ static int show_mask = SHOW_KERNEL | SHOW_USER | SHOW_HV;
36 36
37static int dump_trace = 0; 37static int dump_trace = 0;
38#define dprintf(x...) do { if (dump_trace) printf(x); } while (0) 38#define dprintf(x...) do { if (dump_trace) printf(x); } while (0)
39#define cdprintf(x...) do { if (dump_trace) color_fprintf(stdout, color, x); } while (0)
39 40
40static int verbose; 41static int verbose;
42#define eprintf(x...) do { if (verbose) fprintf(stderr, x); } while (0)
43
41static int full_paths; 44static int full_paths;
42 45
43static unsigned long page_size; 46static unsigned long page_size;
44static unsigned long mmap_window = 32; 47static unsigned long mmap_window = 32;
45 48
49static char default_parent_pattern[] = "^sys_|^do_page_fault";
50static char *parent_pattern = default_parent_pattern;
51static regex_t parent_regex;
52
53static int exclude_other = 1;
54
46struct ip_event { 55struct ip_event {
47 struct perf_event_header header; 56 struct perf_event_header header;
48 __u64 ip; 57 u64 ip;
49 __u32 pid, tid; 58 u32 pid, tid;
50 __u64 period; 59 unsigned char __more_data[];
60};
61
62struct ip_callchain {
63 u64 nr;
64 u64 ips[0];
51}; 65};
52 66
53struct mmap_event { 67struct mmap_event {
54 struct perf_event_header header; 68 struct perf_event_header header;
55 __u32 pid, tid; 69 u32 pid, tid;
56 __u64 start; 70 u64 start;
57 __u64 len; 71 u64 len;
58 __u64 pgoff; 72 u64 pgoff;
59 char filename[PATH_MAX]; 73 char filename[PATH_MAX];
60}; 74};
61 75
62struct comm_event { 76struct comm_event {
63 struct perf_event_header header; 77 struct perf_event_header header;
64 __u32 pid, tid; 78 u32 pid, tid;
65 char comm[16]; 79 char comm[16];
66}; 80};
67 81
68struct fork_event { 82struct fork_event {
69 struct perf_event_header header; 83 struct perf_event_header header;
70 __u32 pid, ppid; 84 u32 pid, ppid;
71}; 85};
72 86
73struct period_event { 87struct period_event {
74 struct perf_event_header header; 88 struct perf_event_header header;
75 __u64 time; 89 u64 time;
76 __u64 id; 90 u64 id;
77 __u64 sample_period; 91 u64 sample_period;
92};
93
94struct lost_event {
95 struct perf_event_header header;
96 u64 id;
97 u64 lost;
78}; 98};
79 99
80typedef union event_union { 100typedef union event_union {
@@ -84,6 +104,7 @@ typedef union event_union {
84 struct comm_event comm; 104 struct comm_event comm;
85 struct fork_event fork; 105 struct fork_event fork;
86 struct period_event period; 106 struct period_event period;
107 struct lost_event lost;
87} event_t; 108} event_t;
88 109
89static LIST_HEAD(dsos); 110static LIST_HEAD(dsos);
@@ -119,15 +140,11 @@ static struct dso *dsos__findnew(const char *name)
119 140
120 nr = dso__load(dso, NULL, verbose); 141 nr = dso__load(dso, NULL, verbose);
121 if (nr < 0) { 142 if (nr < 0) {
122 if (verbose) 143 eprintf("Failed to open: %s\n", name);
123 fprintf(stderr, "Failed to open: %s\n", name);
124 goto out_delete_dso; 144 goto out_delete_dso;
125 } 145 }
126 if (!nr && verbose) { 146 if (!nr)
127 fprintf(stderr, 147 eprintf("No symbols found in: %s, maybe install a debug package?\n", name);
128 "No symbols found in: %s, maybe install a debug package?\n",
129 name);
130 }
131 148
132 dsos__add(dso); 149 dsos__add(dso);
133 150
@@ -146,7 +163,7 @@ static void dsos__fprintf(FILE *fp)
146 dso__fprintf(pos, fp); 163 dso__fprintf(pos, fp);
147} 164}
148 165
149static struct symbol *vdso__find_symbol(struct dso *dso, __u64 ip) 166static struct symbol *vdso__find_symbol(struct dso *dso, u64 ip)
150{ 167{
151 return dso__find_symbol(kernel_dso, ip); 168 return dso__find_symbol(kernel_dso, ip);
152} 169}
@@ -193,19 +210,19 @@ static int strcommon(const char *pathname)
193 210
194struct map { 211struct map {
195 struct list_head node; 212 struct list_head node;
196 __u64 start; 213 u64 start;
197 __u64 end; 214 u64 end;
198 __u64 pgoff; 215 u64 pgoff;
199 __u64 (*map_ip)(struct map *, __u64); 216 u64 (*map_ip)(struct map *, u64);
200 struct dso *dso; 217 struct dso *dso;
201}; 218};
202 219
203static __u64 map__map_ip(struct map *map, __u64 ip) 220static u64 map__map_ip(struct map *map, u64 ip)
204{ 221{
205 return ip - map->start + map->pgoff; 222 return ip - map->start + map->pgoff;
206} 223}
207 224
208static __u64 vdso__map_ip(struct map *map, __u64 ip) 225static u64 vdso__map_ip(struct map *map, u64 ip)
209{ 226{
210 return ip; 227 return ip;
211} 228}
@@ -412,7 +429,7 @@ static int thread__fork(struct thread *self, struct thread *parent)
412 return 0; 429 return 0;
413} 430}
414 431
415static struct map *thread__find_map(struct thread *self, __u64 ip) 432static struct map *thread__find_map(struct thread *self, u64 ip)
416{ 433{
417 struct map *pos; 434 struct map *pos;
418 435
@@ -453,10 +470,11 @@ struct hist_entry {
453 struct map *map; 470 struct map *map;
454 struct dso *dso; 471 struct dso *dso;
455 struct symbol *sym; 472 struct symbol *sym;
456 __u64 ip; 473 struct symbol *parent;
474 u64 ip;
457 char level; 475 char level;
458 476
459 __u64 count; 477 u64 count;
460}; 478};
461 479
462/* 480/*
@@ -473,6 +491,16 @@ struct sort_entry {
473 size_t (*print)(FILE *fp, struct hist_entry *); 491 size_t (*print)(FILE *fp, struct hist_entry *);
474}; 492};
475 493
494static int64_t cmp_null(void *l, void *r)
495{
496 if (!l && !r)
497 return 0;
498 else if (!l)
499 return -1;
500 else
501 return 1;
502}
503
476/* --sort pid */ 504/* --sort pid */
477 505
478static int64_t 506static int64_t
@@ -507,14 +535,8 @@ sort__comm_collapse(struct hist_entry *left, struct hist_entry *right)
507 char *comm_l = left->thread->comm; 535 char *comm_l = left->thread->comm;
508 char *comm_r = right->thread->comm; 536 char *comm_r = right->thread->comm;
509 537
510 if (!comm_l || !comm_r) { 538 if (!comm_l || !comm_r)
511 if (!comm_l && !comm_r) 539 return cmp_null(comm_l, comm_r);
512 return 0;
513 else if (!comm_l)
514 return -1;
515 else
516 return 1;
517 }
518 540
519 return strcmp(comm_l, comm_r); 541 return strcmp(comm_l, comm_r);
520} 542}
@@ -540,14 +562,8 @@ sort__dso_cmp(struct hist_entry *left, struct hist_entry *right)
540 struct dso *dso_l = left->dso; 562 struct dso *dso_l = left->dso;
541 struct dso *dso_r = right->dso; 563 struct dso *dso_r = right->dso;
542 564
543 if (!dso_l || !dso_r) { 565 if (!dso_l || !dso_r)
544 if (!dso_l && !dso_r) 566 return cmp_null(dso_l, dso_r);
545 return 0;
546 else if (!dso_l)
547 return -1;
548 else
549 return 1;
550 }
551 567
552 return strcmp(dso_l->name, dso_r->name); 568 return strcmp(dso_l->name, dso_r->name);
553} 569}
@@ -558,7 +574,7 @@ sort__dso_print(FILE *fp, struct hist_entry *self)
558 if (self->dso) 574 if (self->dso)
559 return fprintf(fp, "%-25s", self->dso->name); 575 return fprintf(fp, "%-25s", self->dso->name);
560 576
561 return fprintf(fp, "%016llx ", (__u64)self->ip); 577 return fprintf(fp, "%016llx ", (u64)self->ip);
562} 578}
563 579
564static struct sort_entry sort_dso = { 580static struct sort_entry sort_dso = {
@@ -572,7 +588,7 @@ static struct sort_entry sort_dso = {
572static int64_t 588static int64_t
573sort__sym_cmp(struct hist_entry *left, struct hist_entry *right) 589sort__sym_cmp(struct hist_entry *left, struct hist_entry *right)
574{ 590{
575 __u64 ip_l, ip_r; 591 u64 ip_l, ip_r;
576 592
577 if (left->sym == right->sym) 593 if (left->sym == right->sym)
578 return 0; 594 return 0;
@@ -589,13 +605,13 @@ sort__sym_print(FILE *fp, struct hist_entry *self)
589 size_t ret = 0; 605 size_t ret = 0;
590 606
591 if (verbose) 607 if (verbose)
592 ret += fprintf(fp, "%#018llx ", (__u64)self->ip); 608 ret += fprintf(fp, "%#018llx ", (u64)self->ip);
593 609
594 if (self->sym) { 610 if (self->sym) {
595 ret += fprintf(fp, "[%c] %s", 611 ret += fprintf(fp, "[%c] %s",
596 self->dso == kernel_dso ? 'k' : '.', self->sym->name); 612 self->dso == kernel_dso ? 'k' : '.', self->sym->name);
597 } else { 613 } else {
598 ret += fprintf(fp, "%#016llx", (__u64)self->ip); 614 ret += fprintf(fp, "%#016llx", (u64)self->ip);
599 } 615 }
600 616
601 return ret; 617 return ret;
@@ -607,7 +623,38 @@ static struct sort_entry sort_sym = {
607 .print = sort__sym_print, 623 .print = sort__sym_print,
608}; 624};
609 625
626/* --sort parent */
627
628static int64_t
629sort__parent_cmp(struct hist_entry *left, struct hist_entry *right)
630{
631 struct symbol *sym_l = left->parent;
632 struct symbol *sym_r = right->parent;
633
634 if (!sym_l || !sym_r)
635 return cmp_null(sym_l, sym_r);
636
637 return strcmp(sym_l->name, sym_r->name);
638}
639
640static size_t
641sort__parent_print(FILE *fp, struct hist_entry *self)
642{
643 size_t ret = 0;
644
645 ret += fprintf(fp, "%-20s", self->parent ? self->parent->name : "[other]");
646
647 return ret;
648}
649
650static struct sort_entry sort_parent = {
651 .header = "Parent symbol ",
652 .cmp = sort__parent_cmp,
653 .print = sort__parent_print,
654};
655
610static int sort__need_collapse = 0; 656static int sort__need_collapse = 0;
657static int sort__has_parent = 0;
611 658
612struct sort_dimension { 659struct sort_dimension {
613 char *name; 660 char *name;
@@ -620,6 +667,7 @@ static struct sort_dimension sort_dimensions[] = {
620 { .name = "comm", .entry = &sort_comm, }, 667 { .name = "comm", .entry = &sort_comm, },
621 { .name = "dso", .entry = &sort_dso, }, 668 { .name = "dso", .entry = &sort_dso, },
622 { .name = "symbol", .entry = &sort_sym, }, 669 { .name = "symbol", .entry = &sort_sym, },
670 { .name = "parent", .entry = &sort_parent, },
623}; 671};
624 672
625static LIST_HEAD(hist_entry__sort_list); 673static LIST_HEAD(hist_entry__sort_list);
@@ -640,6 +688,19 @@ static int sort_dimension__add(char *tok)
640 if (sd->entry->collapse) 688 if (sd->entry->collapse)
641 sort__need_collapse = 1; 689 sort__need_collapse = 1;
642 690
691 if (sd->entry == &sort_parent) {
692 int ret = regcomp(&parent_regex, parent_pattern, REG_EXTENDED);
693 if (ret) {
694 char err[BUFSIZ];
695
696 regerror(ret, &parent_regex, err, sizeof(err));
697 fprintf(stderr, "Invalid regex: %s\n%s",
698 parent_pattern, err);
699 exit(-1);
700 }
701 sort__has_parent = 1;
702 }
703
643 list_add_tail(&sd->entry->list, &hist_entry__sort_list); 704 list_add_tail(&sd->entry->list, &hist_entry__sort_list);
644 sd->taken = 1; 705 sd->taken = 1;
645 706
@@ -684,11 +745,14 @@ hist_entry__collapse(struct hist_entry *left, struct hist_entry *right)
684} 745}
685 746
686static size_t 747static size_t
687hist_entry__fprintf(FILE *fp, struct hist_entry *self, __u64 total_samples) 748hist_entry__fprintf(FILE *fp, struct hist_entry *self, u64 total_samples)
688{ 749{
689 struct sort_entry *se; 750 struct sort_entry *se;
690 size_t ret; 751 size_t ret;
691 752
753 if (exclude_other && !self->parent)
754 return 0;
755
692 if (total_samples) { 756 if (total_samples) {
693 double percent = self->count * 100.0 / total_samples; 757 double percent = self->count * 100.0 / total_samples;
694 char *color = PERF_COLOR_NORMAL; 758 char *color = PERF_COLOR_NORMAL;
@@ -711,6 +775,9 @@ hist_entry__fprintf(FILE *fp, struct hist_entry *self, __u64 total_samples)
711 ret = fprintf(fp, "%12Ld ", self->count); 775 ret = fprintf(fp, "%12Ld ", self->count);
712 776
713 list_for_each_entry(se, &hist_entry__sort_list, list) { 777 list_for_each_entry(se, &hist_entry__sort_list, list) {
778 if (exclude_other && (se == &sort_parent))
779 continue;
780
714 fprintf(fp, " "); 781 fprintf(fp, " ");
715 ret += se->print(fp, self); 782 ret += se->print(fp, self);
716 } 783 }
@@ -721,12 +788,72 @@ hist_entry__fprintf(FILE *fp, struct hist_entry *self, __u64 total_samples)
721} 788}
722 789
723/* 790/*
791 *
792 */
793
794static struct symbol *
795resolve_symbol(struct thread *thread, struct map **mapp,
796 struct dso **dsop, u64 *ipp)
797{
798 struct dso *dso = dsop ? *dsop : NULL;
799 struct map *map = mapp ? *mapp : NULL;
800 uint64_t ip = *ipp;
801
802 if (!thread)
803 return NULL;
804
805 if (dso)
806 goto got_dso;
807
808 if (map)
809 goto got_map;
810
811 map = thread__find_map(thread, ip);
812 if (map != NULL) {
813 if (mapp)
814 *mapp = map;
815got_map:
816 ip = map->map_ip(map, ip);
817 *ipp = ip;
818
819 dso = map->dso;
820 } else {
821 /*
822 * If this is outside of all known maps,
823 * and is a negative address, try to look it
824 * up in the kernel dso, as it might be a
825 * vsyscall (which executes in user-mode):
826 */
827 if ((long long)ip < 0)
828 dso = kernel_dso;
829 }
830 dprintf(" ...... dso: %s\n", dso ? dso->name : "<not found>");
831
832 if (dsop)
833 *dsop = dso;
834
835 if (!dso)
836 return NULL;
837got_dso:
838 return dso->find_symbol(dso, ip);
839}
840
841static int call__match(struct symbol *sym)
842{
843 if (sym->name && !regexec(&parent_regex, sym->name, 0, NULL, 0))
844 return 1;
845
846 return 0;
847}
848
849/*
724 * collect histogram counts 850 * collect histogram counts
725 */ 851 */
726 852
727static int 853static int
728hist_entry__add(struct thread *thread, struct map *map, struct dso *dso, 854hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
729 struct symbol *sym, __u64 ip, char level, __u64 count) 855 struct symbol *sym, u64 ip, struct ip_callchain *chain,
856 char level, u64 count)
730{ 857{
731 struct rb_node **p = &hist.rb_node; 858 struct rb_node **p = &hist.rb_node;
732 struct rb_node *parent = NULL; 859 struct rb_node *parent = NULL;
@@ -739,9 +866,41 @@ hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
739 .ip = ip, 866 .ip = ip,
740 .level = level, 867 .level = level,
741 .count = count, 868 .count = count,
869 .parent = NULL,
742 }; 870 };
743 int cmp; 871 int cmp;
744 872
873 if (sort__has_parent && chain) {
874 u64 context = PERF_CONTEXT_MAX;
875 int i;
876
877 for (i = 0; i < chain->nr; i++) {
878 u64 ip = chain->ips[i];
879 struct dso *dso = NULL;
880 struct symbol *sym;
881
882 if (ip >= PERF_CONTEXT_MAX) {
883 context = ip;
884 continue;
885 }
886
887 switch (context) {
888 case PERF_CONTEXT_KERNEL:
889 dso = kernel_dso;
890 break;
891 default:
892 break;
893 }
894
895 sym = resolve_symbol(thread, NULL, &dso, &ip);
896
897 if (sym && call__match(sym)) {
898 entry.parent = sym;
899 break;
900 }
901 }
902 }
903
745 while (*p != NULL) { 904 while (*p != NULL) {
746 parent = *p; 905 parent = *p;
747 he = rb_entry(parent, struct hist_entry, rb_node); 906 he = rb_entry(parent, struct hist_entry, rb_node);
@@ -873,7 +1032,7 @@ static void output__resort(void)
873 } 1032 }
874} 1033}
875 1034
876static size_t output__fprintf(FILE *fp, __u64 total_samples) 1035static size_t output__fprintf(FILE *fp, u64 total_samples)
877{ 1036{
878 struct hist_entry *pos; 1037 struct hist_entry *pos;
879 struct sort_entry *se; 1038 struct sort_entry *se;
@@ -882,18 +1041,24 @@ static size_t output__fprintf(FILE *fp, __u64 total_samples)
882 1041
883 fprintf(fp, "\n"); 1042 fprintf(fp, "\n");
884 fprintf(fp, "#\n"); 1043 fprintf(fp, "#\n");
885 fprintf(fp, "# (%Ld samples)\n", (__u64)total_samples); 1044 fprintf(fp, "# (%Ld samples)\n", (u64)total_samples);
886 fprintf(fp, "#\n"); 1045 fprintf(fp, "#\n");
887 1046
888 fprintf(fp, "# Overhead"); 1047 fprintf(fp, "# Overhead");
889 list_for_each_entry(se, &hist_entry__sort_list, list) 1048 list_for_each_entry(se, &hist_entry__sort_list, list) {
1049 if (exclude_other && (se == &sort_parent))
1050 continue;
890 fprintf(fp, " %s", se->header); 1051 fprintf(fp, " %s", se->header);
1052 }
891 fprintf(fp, "\n"); 1053 fprintf(fp, "\n");
892 1054
893 fprintf(fp, "# ........"); 1055 fprintf(fp, "# ........");
894 list_for_each_entry(se, &hist_entry__sort_list, list) { 1056 list_for_each_entry(se, &hist_entry__sort_list, list) {
895 int i; 1057 int i;
896 1058
1059 if (exclude_other && (se == &sort_parent))
1060 continue;
1061
897 fprintf(fp, " "); 1062 fprintf(fp, " ");
898 for (i = 0; i < strlen(se->header); i++) 1063 for (i = 0; i < strlen(se->header); i++)
899 fprintf(fp, "."); 1064 fprintf(fp, ".");
@@ -907,7 +1072,8 @@ static size_t output__fprintf(FILE *fp, __u64 total_samples)
907 ret += hist_entry__fprintf(fp, pos, total_samples); 1072 ret += hist_entry__fprintf(fp, pos, total_samples);
908 } 1073 }
909 1074
910 if (!strcmp(sort_order, default_sort_order)) { 1075 if (sort_order == default_sort_order &&
1076 parent_pattern == default_parent_pattern) {
911 fprintf(fp, "#\n"); 1077 fprintf(fp, "#\n");
912 fprintf(fp, "# (For more details, try: perf report --sort comm,dso,symbol)\n"); 1078 fprintf(fp, "# (For more details, try: perf report --sort comm,dso,symbol)\n");
913 fprintf(fp, "#\n"); 1079 fprintf(fp, "#\n");
@@ -932,7 +1098,21 @@ static unsigned long total = 0,
932 total_mmap = 0, 1098 total_mmap = 0,
933 total_comm = 0, 1099 total_comm = 0,
934 total_fork = 0, 1100 total_fork = 0,
935 total_unknown = 0; 1101 total_unknown = 0,
1102 total_lost = 0;
1103
1104static int validate_chain(struct ip_callchain *chain, event_t *event)
1105{
1106 unsigned int chain_size;
1107
1108 chain_size = event->header.size;
1109 chain_size -= (unsigned long)&event->ip.__more_data - (unsigned long)event;
1110
1111 if (chain->nr*sizeof(u64) > chain_size)
1112 return -1;
1113
1114 return 0;
1115}
936 1116
937static int 1117static int
938process_overflow_event(event_t *event, unsigned long offset, unsigned long head) 1118process_overflow_event(event_t *event, unsigned long offset, unsigned long head)
@@ -941,12 +1121,16 @@ process_overflow_event(event_t *event, unsigned long offset, unsigned long head)
941 int show = 0; 1121 int show = 0;
942 struct dso *dso = NULL; 1122 struct dso *dso = NULL;
943 struct thread *thread = threads__findnew(event->ip.pid); 1123 struct thread *thread = threads__findnew(event->ip.pid);
944 __u64 ip = event->ip.ip; 1124 u64 ip = event->ip.ip;
945 __u64 period = 1; 1125 u64 period = 1;
946 struct map *map = NULL; 1126 struct map *map = NULL;
1127 void *more_data = event->ip.__more_data;
1128 struct ip_callchain *chain = NULL;
947 1129
948 if (event->header.type & PERF_SAMPLE_PERIOD) 1130 if (event->header.type & PERF_SAMPLE_PERIOD) {
949 period = event->ip.period; 1131 period = *(u64 *)more_data;
1132 more_data += sizeof(u64);
1133 }
950 1134
951 dprintf("%p [%p]: PERF_EVENT (IP, %d): %d: %p period: %Ld\n", 1135 dprintf("%p [%p]: PERF_EVENT (IP, %d): %d: %p period: %Ld\n",
952 (void *)(offset + head), 1136 (void *)(offset + head),
@@ -956,10 +1140,28 @@ process_overflow_event(event_t *event, unsigned long offset, unsigned long head)
956 (void *)(long)ip, 1140 (void *)(long)ip,
957 (long long)period); 1141 (long long)period);
958 1142
1143 if (event->header.type & PERF_SAMPLE_CALLCHAIN) {
1144 int i;
1145
1146 chain = (void *)more_data;
1147
1148 dprintf("... chain: nr:%Lu\n", chain->nr);
1149
1150 if (validate_chain(chain, event) < 0) {
1151 eprintf("call-chain problem with event, skipping it.\n");
1152 return 0;
1153 }
1154
1155 if (dump_trace) {
1156 for (i = 0; i < chain->nr; i++)
1157 dprintf("..... %2d: %016Lx\n", i, chain->ips[i]);
1158 }
1159 }
1160
959 dprintf(" ... thread: %s:%d\n", thread->comm, thread->pid); 1161 dprintf(" ... thread: %s:%d\n", thread->comm, thread->pid);
960 1162
961 if (thread == NULL) { 1163 if (thread == NULL) {
962 fprintf(stderr, "problem processing %d event, skipping it.\n", 1164 eprintf("problem processing %d event, skipping it.\n",
963 event->header.type); 1165 event->header.type);
964 return -1; 1166 return -1;
965 } 1167 }
@@ -977,22 +1179,6 @@ process_overflow_event(event_t *event, unsigned long offset, unsigned long head)
977 show = SHOW_USER; 1179 show = SHOW_USER;
978 level = '.'; 1180 level = '.';
979 1181
980 map = thread__find_map(thread, ip);
981 if (map != NULL) {
982 ip = map->map_ip(map, ip);
983 dso = map->dso;
984 } else {
985 /*
986 * If this is outside of all known maps,
987 * and is a negative address, try to look it
988 * up in the kernel dso, as it might be a
989 * vsyscall (which executes in user-mode):
990 */
991 if ((long long)ip < 0)
992 dso = kernel_dso;
993 }
994 dprintf(" ...... dso: %s\n", dso ? dso->name : "<not found>");
995
996 } else { 1182 } else {
997 show = SHOW_HV; 1183 show = SHOW_HV;
998 level = 'H'; 1184 level = 'H';
@@ -1000,14 +1186,10 @@ process_overflow_event(event_t *event, unsigned long offset, unsigned long head)
1000 } 1186 }
1001 1187
1002 if (show & show_mask) { 1188 if (show & show_mask) {
1003 struct symbol *sym = NULL; 1189 struct symbol *sym = resolve_symbol(thread, &map, &dso, &ip);
1004
1005 if (dso)
1006 sym = dso->find_symbol(dso, ip);
1007 1190
1008 if (hist_entry__add(thread, map, dso, sym, ip, level, period)) { 1191 if (hist_entry__add(thread, map, dso, sym, ip, chain, level, period)) {
1009 fprintf(stderr, 1192 eprintf("problem incrementing symbol count, skipping event\n");
1010 "problem incrementing symbol count, skipping event\n");
1011 return -1; 1193 return -1;
1012 } 1194 }
1013 } 1195 }
@@ -1096,8 +1278,60 @@ process_period_event(event_t *event, unsigned long offset, unsigned long head)
1096} 1278}
1097 1279
1098static int 1280static int
1281process_lost_event(event_t *event, unsigned long offset, unsigned long head)
1282{
1283 dprintf("%p [%p]: PERF_EVENT_LOST: id:%Ld: lost:%Ld\n",
1284 (void *)(offset + head),
1285 (void *)(long)(event->header.size),
1286 event->lost.id,
1287 event->lost.lost);
1288
1289 total_lost += event->lost.lost;
1290
1291 return 0;
1292}
1293
1294static void trace_event(event_t *event)
1295{
1296 unsigned char *raw_event = (void *)event;
1297 char *color = PERF_COLOR_BLUE;
1298 int i, j;
1299
1300 if (!dump_trace)
1301 return;
1302
1303 dprintf(".");
1304 cdprintf("\n. ... raw event: size %d bytes\n", event->header.size);
1305
1306 for (i = 0; i < event->header.size; i++) {
1307 if ((i & 15) == 0) {
1308 dprintf(".");
1309 cdprintf(" %04x: ", i);
1310 }
1311
1312 cdprintf(" %02x", raw_event[i]);
1313
1314 if (((i & 15) == 15) || i == event->header.size-1) {
1315 cdprintf(" ");
1316 for (j = 0; j < 15-(i & 15); j++)
1317 cdprintf(" ");
1318 for (j = 0; j < (i & 15); j++) {
1319 if (isprint(raw_event[i-15+j]))
1320 cdprintf("%c", raw_event[i-15+j]);
1321 else
1322 cdprintf(".");
1323 }
1324 cdprintf("\n");
1325 }
1326 }
1327 dprintf(".\n");
1328}
1329
1330static int
1099process_event(event_t *event, unsigned long offset, unsigned long head) 1331process_event(event_t *event, unsigned long offset, unsigned long head)
1100{ 1332{
1333 trace_event(event);
1334
1101 if (event->header.misc & PERF_EVENT_MISC_OVERFLOW) 1335 if (event->header.misc & PERF_EVENT_MISC_OVERFLOW)
1102 return process_overflow_event(event, offset, head); 1336 return process_overflow_event(event, offset, head);
1103 1337
@@ -1113,6 +1347,10 @@ process_event(event_t *event, unsigned long offset, unsigned long head)
1113 1347
1114 case PERF_EVENT_PERIOD: 1348 case PERF_EVENT_PERIOD:
1115 return process_period_event(event, offset, head); 1349 return process_period_event(event, offset, head);
1350
1351 case PERF_EVENT_LOST:
1352 return process_lost_event(event, offset, head);
1353
1116 /* 1354 /*
1117 * We dont process them right now but they are fine: 1355 * We dont process them right now but they are fine:
1118 */ 1356 */
@@ -1128,11 +1366,13 @@ process_event(event_t *event, unsigned long offset, unsigned long head)
1128 return 0; 1366 return 0;
1129} 1367}
1130 1368
1369static struct perf_file_header file_header;
1370
1131static int __cmd_report(void) 1371static int __cmd_report(void)
1132{ 1372{
1133 int ret, rc = EXIT_FAILURE; 1373 int ret, rc = EXIT_FAILURE;
1134 unsigned long offset = 0; 1374 unsigned long offset = 0;
1135 unsigned long head = 0; 1375 unsigned long head = sizeof(file_header);
1136 struct stat stat; 1376 struct stat stat;
1137 event_t *event; 1377 event_t *event;
1138 uint32_t size; 1378 uint32_t size;
@@ -1160,6 +1400,17 @@ static int __cmd_report(void)
1160 exit(0); 1400 exit(0);
1161 } 1401 }
1162 1402
1403 if (read(input, &file_header, sizeof(file_header)) == -1) {
1404 perror("failed to read file headers");
1405 exit(-1);
1406 }
1407
1408 if (sort__has_parent &&
1409 !(file_header.sample_type & PERF_SAMPLE_CALLCHAIN)) {
1410 fprintf(stderr, "selected --sort parent, but no callchain data\n");
1411 exit(-1);
1412 }
1413
1163 if (load_kernel() < 0) { 1414 if (load_kernel() < 0) {
1164 perror("failed to load kernel symbols"); 1415 perror("failed to load kernel symbols");
1165 return EXIT_FAILURE; 1416 return EXIT_FAILURE;
@@ -1204,7 +1455,7 @@ more:
1204 1455
1205 size = event->header.size; 1456 size = event->header.size;
1206 1457
1207 dprintf("%p [%p]: event: %d\n", 1458 dprintf("\n%p [%p]: event: %d\n",
1208 (void *)(offset + head), 1459 (void *)(offset + head),
1209 (void *)(long)event->header.size, 1460 (void *)(long)event->header.size,
1210 event->header.type); 1461 event->header.type);
@@ -1231,9 +1482,13 @@ more:
1231 1482
1232 head += size; 1483 head += size;
1233 1484
1485 if (offset + head >= sizeof(file_header) + file_header.data_size)
1486 goto done;
1487
1234 if (offset + head < stat.st_size) 1488 if (offset + head < stat.st_size)
1235 goto more; 1489 goto more;
1236 1490
1491done:
1237 rc = EXIT_SUCCESS; 1492 rc = EXIT_SUCCESS;
1238 close(input); 1493 close(input);
1239 1494
@@ -1241,6 +1496,7 @@ more:
1241 dprintf(" mmap events: %10ld\n", total_mmap); 1496 dprintf(" mmap events: %10ld\n", total_mmap);
1242 dprintf(" comm events: %10ld\n", total_comm); 1497 dprintf(" comm events: %10ld\n", total_comm);
1243 dprintf(" fork events: %10ld\n", total_fork); 1498 dprintf(" fork events: %10ld\n", total_fork);
1499 dprintf(" lost events: %10ld\n", total_lost);
1244 dprintf(" unknown events: %10ld\n", total_unknown); 1500 dprintf(" unknown events: %10ld\n", total_unknown);
1245 1501
1246 if (dump_trace) 1502 if (dump_trace)
@@ -1273,9 +1529,13 @@ static const struct option options[] = {
1273 "dump raw trace in ASCII"), 1529 "dump raw trace in ASCII"),
1274 OPT_STRING('k', "vmlinux", &vmlinux, "file", "vmlinux pathname"), 1530 OPT_STRING('k', "vmlinux", &vmlinux, "file", "vmlinux pathname"),
1275 OPT_STRING('s', "sort", &sort_order, "key[,key2...]", 1531 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1276 "sort by key(s): pid, comm, dso, symbol. Default: pid,symbol"), 1532 "sort by key(s): pid, comm, dso, symbol, parent"),
1277 OPT_BOOLEAN('P', "full-paths", &full_paths, 1533 OPT_BOOLEAN('P', "full-paths", &full_paths,
1278 "Don't shorten the pathnames taking into account the cwd"), 1534 "Don't shorten the pathnames taking into account the cwd"),
1535 OPT_STRING('p', "parent", &parent_pattern, "regex",
1536 "regex filter to identify parent, see: '--sort parent'"),
1537 OPT_BOOLEAN('x', "exclude-other", &exclude_other,
1538 "Only display entries with parent-match"),
1279 OPT_END() 1539 OPT_END()
1280}; 1540};
1281 1541
@@ -1304,6 +1564,11 @@ int cmd_report(int argc, const char **argv, const char *prefix)
1304 1564
1305 setup_sorting(); 1565 setup_sorting();
1306 1566
1567 if (parent_pattern != default_parent_pattern)
1568 sort_dimension__add("parent");
1569 else
1570 exclude_other = 0;
1571
1307 /* 1572 /*
1308 * Any (unrecognized) arguments left? 1573 * Any (unrecognized) arguments left?
1309 */ 1574 */
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index c43e4a97dc42..6d3eeac1ea25 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -43,6 +43,7 @@
43#include "util/parse-events.h" 43#include "util/parse-events.h"
44 44
45#include <sys/prctl.h> 45#include <sys/prctl.h>
46#include <math.h>
46 47
47static struct perf_counter_attr default_attrs[MAX_COUNTERS] = { 48static struct perf_counter_attr default_attrs[MAX_COUNTERS] = {
48 49
@@ -79,12 +80,34 @@ static const unsigned int default_count[] = {
79 10000, 80 10000,
80}; 81};
81 82
82static __u64 event_res[MAX_COUNTERS][3]; 83#define MAX_RUN 100
83static __u64 event_scaled[MAX_COUNTERS];
84 84
85static __u64 runtime_nsecs; 85static int run_count = 1;
86static __u64 walltime_nsecs; 86static int run_idx = 0;
87static __u64 runtime_cycles; 87
88static u64 event_res[MAX_RUN][MAX_COUNTERS][3];
89static u64 event_scaled[MAX_RUN][MAX_COUNTERS];
90
91//static u64 event_hist[MAX_RUN][MAX_COUNTERS][3];
92
93
94static u64 runtime_nsecs[MAX_RUN];
95static u64 walltime_nsecs[MAX_RUN];
96static u64 runtime_cycles[MAX_RUN];
97
98static u64 event_res_avg[MAX_COUNTERS][3];
99static u64 event_res_noise[MAX_COUNTERS][3];
100
101static u64 event_scaled_avg[MAX_COUNTERS];
102
103static u64 runtime_nsecs_avg;
104static u64 runtime_nsecs_noise;
105
106static u64 walltime_nsecs_avg;
107static u64 walltime_nsecs_noise;
108
109static u64 runtime_cycles_avg;
110static u64 runtime_cycles_noise;
88 111
89static void create_perf_stat_counter(int counter) 112static void create_perf_stat_counter(int counter)
90{ 113{
@@ -135,12 +158,12 @@ static inline int nsec_counter(int counter)
135 */ 158 */
136static void read_counter(int counter) 159static void read_counter(int counter)
137{ 160{
138 __u64 *count, single_count[3]; 161 u64 *count, single_count[3];
139 ssize_t res; 162 ssize_t res;
140 int cpu, nv; 163 int cpu, nv;
141 int scaled; 164 int scaled;
142 165
143 count = event_res[counter]; 166 count = event_res[run_idx][counter];
144 167
145 count[0] = count[1] = count[2] = 0; 168 count[0] = count[1] = count[2] = 0;
146 169
@@ -149,8 +172,10 @@ static void read_counter(int counter)
149 if (fd[cpu][counter] < 0) 172 if (fd[cpu][counter] < 0)
150 continue; 173 continue;
151 174
152 res = read(fd[cpu][counter], single_count, nv * sizeof(__u64)); 175 res = read(fd[cpu][counter], single_count, nv * sizeof(u64));
153 assert(res == nv * sizeof(__u64)); 176 assert(res == nv * sizeof(u64));
177 close(fd[cpu][counter]);
178 fd[cpu][counter] = -1;
154 179
155 count[0] += single_count[0]; 180 count[0] += single_count[0];
156 if (scale) { 181 if (scale) {
@@ -162,13 +187,13 @@ static void read_counter(int counter)
162 scaled = 0; 187 scaled = 0;
163 if (scale) { 188 if (scale) {
164 if (count[2] == 0) { 189 if (count[2] == 0) {
165 event_scaled[counter] = -1; 190 event_scaled[run_idx][counter] = -1;
166 count[0] = 0; 191 count[0] = 0;
167 return; 192 return;
168 } 193 }
169 194
170 if (count[2] < count[1]) { 195 if (count[2] < count[1]) {
171 event_scaled[counter] = 1; 196 event_scaled[run_idx][counter] = 1;
172 count[0] = (unsigned long long) 197 count[0] = (unsigned long long)
173 ((double)count[0] * count[1] / count[2] + 0.5); 198 ((double)count[0] * count[1] / count[2] + 0.5);
174 } 199 }
@@ -178,10 +203,94 @@ static void read_counter(int counter)
178 */ 203 */
179 if (attrs[counter].type == PERF_TYPE_SOFTWARE && 204 if (attrs[counter].type == PERF_TYPE_SOFTWARE &&
180 attrs[counter].config == PERF_COUNT_SW_TASK_CLOCK) 205 attrs[counter].config == PERF_COUNT_SW_TASK_CLOCK)
181 runtime_nsecs = count[0]; 206 runtime_nsecs[run_idx] = count[0];
182 if (attrs[counter].type == PERF_TYPE_HARDWARE && 207 if (attrs[counter].type == PERF_TYPE_HARDWARE &&
183 attrs[counter].config == PERF_COUNT_HW_CPU_CYCLES) 208 attrs[counter].config == PERF_COUNT_HW_CPU_CYCLES)
184 runtime_cycles = count[0]; 209 runtime_cycles[run_idx] = count[0];
210}
211
212static int run_perf_stat(int argc, const char **argv)
213{
214 unsigned long long t0, t1;
215 int status = 0;
216 int counter;
217 int pid;
218
219 if (!system_wide)
220 nr_cpus = 1;
221
222 for (counter = 0; counter < nr_counters; counter++)
223 create_perf_stat_counter(counter);
224
225 /*
226 * Enable counters and exec the command:
227 */
228 t0 = rdclock();
229 prctl(PR_TASK_PERF_COUNTERS_ENABLE);
230
231 if ((pid = fork()) < 0)
232 perror("failed to fork");
233
234 if (!pid) {
235 if (execvp(argv[0], (char **)argv)) {
236 perror(argv[0]);
237 exit(-1);
238 }
239 }
240
241 wait(&status);
242
243 prctl(PR_TASK_PERF_COUNTERS_DISABLE);
244 t1 = rdclock();
245
246 walltime_nsecs[run_idx] = t1 - t0;
247
248 for (counter = 0; counter < nr_counters; counter++)
249 read_counter(counter);
250
251 return WEXITSTATUS(status);
252}
253
254static void print_noise(u64 *count, u64 *noise)
255{
256 if (run_count > 1)
257 fprintf(stderr, " ( +- %7.3f%% )",
258 (double)noise[0]/(count[0]+1)*100.0);
259}
260
261static void nsec_printout(int counter, u64 *count, u64 *noise)
262{
263 double msecs = (double)count[0] / 1000000;
264
265 fprintf(stderr, " %14.6f %-20s", msecs, event_name(counter));
266
267 if (attrs[counter].type == PERF_TYPE_SOFTWARE &&
268 attrs[counter].config == PERF_COUNT_SW_TASK_CLOCK) {
269
270 if (walltime_nsecs_avg)
271 fprintf(stderr, " # %10.3f CPUs ",
272 (double)count[0] / (double)walltime_nsecs_avg);
273 }
274 print_noise(count, noise);
275}
276
277static void abs_printout(int counter, u64 *count, u64 *noise)
278{
279 fprintf(stderr, " %14Ld %-20s", count[0], event_name(counter));
280
281 if (runtime_cycles_avg &&
282 attrs[counter].type == PERF_TYPE_HARDWARE &&
283 attrs[counter].config == PERF_COUNT_HW_INSTRUCTIONS) {
284
285 fprintf(stderr, " # %10.3f IPC ",
286 (double)count[0] / (double)runtime_cycles_avg);
287 } else {
288 if (runtime_nsecs_avg) {
289 fprintf(stderr, " # %10.3f M/sec",
290 (double)count[0]/runtime_nsecs_avg*1000.0);
291 }
292 }
293 print_noise(count, noise);
185} 294}
186 295
187/* 296/*
@@ -189,11 +298,12 @@ static void read_counter(int counter)
189 */ 298 */
190static void print_counter(int counter) 299static void print_counter(int counter)
191{ 300{
192 __u64 *count; 301 u64 *count, *noise;
193 int scaled; 302 int scaled;
194 303
195 count = event_res[counter]; 304 count = event_res_avg[counter];
196 scaled = event_scaled[counter]; 305 noise = event_res_noise[counter];
306 scaled = event_scaled_avg[counter];
197 307
198 if (scaled == -1) { 308 if (scaled == -1) {
199 fprintf(stderr, " %14s %-20s\n", 309 fprintf(stderr, " %14s %-20s\n",
@@ -201,75 +311,107 @@ static void print_counter(int counter)
201 return; 311 return;
202 } 312 }
203 313
204 if (nsec_counter(counter)) { 314 if (nsec_counter(counter))
205 double msecs = (double)count[0] / 1000000; 315 nsec_printout(counter, count, noise);
206 316 else
207 fprintf(stderr, " %14.6f %-20s", 317 abs_printout(counter, count, noise);
208 msecs, event_name(counter));
209 if (attrs[counter].type == PERF_TYPE_SOFTWARE &&
210 attrs[counter].config == PERF_COUNT_SW_TASK_CLOCK) {
211 318
212 if (walltime_nsecs)
213 fprintf(stderr, " # %11.3f CPU utilization factor",
214 (double)count[0] / (double)walltime_nsecs);
215 }
216 } else {
217 fprintf(stderr, " %14Ld %-20s",
218 count[0], event_name(counter));
219 if (runtime_nsecs)
220 fprintf(stderr, " # %11.3f M/sec",
221 (double)count[0]/runtime_nsecs*1000.0);
222 if (runtime_cycles &&
223 attrs[counter].type == PERF_TYPE_HARDWARE &&
224 attrs[counter].config == PERF_COUNT_HW_INSTRUCTIONS) {
225
226 fprintf(stderr, " # %1.3f per cycle",
227 (double)count[0] / (double)runtime_cycles);
228 }
229 }
230 if (scaled) 319 if (scaled)
231 fprintf(stderr, " (scaled from %.2f%%)", 320 fprintf(stderr, " (scaled from %.2f%%)",
232 (double) count[2] / count[1] * 100); 321 (double) count[2] / count[1] * 100);
322
233 fprintf(stderr, "\n"); 323 fprintf(stderr, "\n");
234} 324}
235 325
236static int do_perf_stat(int argc, const char **argv) 326/*
327 * normalize_noise noise values down to stddev:
328 */
329static void normalize_noise(u64 *val)
237{ 330{
238 unsigned long long t0, t1; 331 double res;
239 int counter;
240 int status;
241 int pid;
242 int i;
243 332
244 if (!system_wide) 333 res = (double)*val / (run_count * sqrt((double)run_count));
245 nr_cpus = 1;
246 334
247 for (counter = 0; counter < nr_counters; counter++) 335 *val = (u64)res;
248 create_perf_stat_counter(counter); 336}
249 337
250 /* 338static void update_avg(const char *name, int idx, u64 *avg, u64 *val)
251 * Enable counters and exec the command: 339{
252 */ 340 *avg += *val;
253 t0 = rdclock();
254 prctl(PR_TASK_PERF_COUNTERS_ENABLE);
255 341
256 if ((pid = fork()) < 0) 342 if (verbose > 1)
257 perror("failed to fork"); 343 fprintf(stderr, "debug: %20s[%d]: %Ld\n", name, idx, *val);
344}
345/*
346 * Calculate the averages and noises:
347 */
348static void calc_avg(void)
349{
350 int i, j;
351
352 if (verbose > 1)
353 fprintf(stderr, "\n");
354
355 for (i = 0; i < run_count; i++) {
356 update_avg("runtime", 0, &runtime_nsecs_avg, runtime_nsecs + i);
357 update_avg("walltime", 0, &walltime_nsecs_avg, walltime_nsecs + i);
358 update_avg("runtime_cycles", 0, &runtime_cycles_avg, runtime_cycles + i);
359
360 for (j = 0; j < nr_counters; j++) {
361 update_avg("counter/0", j,
362 event_res_avg[j]+0, event_res[i][j]+0);
363 update_avg("counter/1", j,
364 event_res_avg[j]+1, event_res[i][j]+1);
365 update_avg("counter/2", j,
366 event_res_avg[j]+2, event_res[i][j]+2);
367 update_avg("scaled", j,
368 event_scaled_avg + j, event_scaled[i]+j);
369 }
370 }
371 runtime_nsecs_avg /= run_count;
372 walltime_nsecs_avg /= run_count;
373 runtime_cycles_avg /= run_count;
374
375 for (j = 0; j < nr_counters; j++) {
376 event_res_avg[j][0] /= run_count;
377 event_res_avg[j][1] /= run_count;
378 event_res_avg[j][2] /= run_count;
379 }
258 380
259 if (!pid) { 381 for (i = 0; i < run_count; i++) {
260 if (execvp(argv[0], (char **)argv)) { 382 runtime_nsecs_noise +=
261 perror(argv[0]); 383 abs((s64)(runtime_nsecs[i] - runtime_nsecs_avg));
262 exit(-1); 384 walltime_nsecs_noise +=
385 abs((s64)(walltime_nsecs[i] - walltime_nsecs_avg));
386 runtime_cycles_noise +=
387 abs((s64)(runtime_cycles[i] - runtime_cycles_avg));
388
389 for (j = 0; j < nr_counters; j++) {
390 event_res_noise[j][0] +=
391 abs((s64)(event_res[i][j][0] - event_res_avg[j][0]));
392 event_res_noise[j][1] +=
393 abs((s64)(event_res[i][j][1] - event_res_avg[j][1]));
394 event_res_noise[j][2] +=
395 abs((s64)(event_res[i][j][2] - event_res_avg[j][2]));
263 } 396 }
264 } 397 }
265 398
266 while (wait(&status) >= 0) 399 normalize_noise(&runtime_nsecs_noise);
267 ; 400 normalize_noise(&walltime_nsecs_noise);
401 normalize_noise(&runtime_cycles_noise);
268 402
269 prctl(PR_TASK_PERF_COUNTERS_DISABLE); 403 for (j = 0; j < nr_counters; j++) {
270 t1 = rdclock(); 404 normalize_noise(&event_res_noise[j][0]);
405 normalize_noise(&event_res_noise[j][1]);
406 normalize_noise(&event_res_noise[j][2]);
407 }
408}
271 409
272 walltime_nsecs = t1 - t0; 410static void print_stat(int argc, const char **argv)
411{
412 int i, counter;
413
414 calc_avg();
273 415
274 fflush(stdout); 416 fflush(stdout);
275 417
@@ -279,22 +421,19 @@ static int do_perf_stat(int argc, const char **argv)
279 for (i = 1; i < argc; i++) 421 for (i = 1; i < argc; i++)
280 fprintf(stderr, " %s", argv[i]); 422 fprintf(stderr, " %s", argv[i]);
281 423
282 fprintf(stderr, "\':\n"); 424 fprintf(stderr, "\'");
283 fprintf(stderr, "\n"); 425 if (run_count > 1)
284 426 fprintf(stderr, " (%d runs)", run_count);
285 for (counter = 0; counter < nr_counters; counter++) 427 fprintf(stderr, ":\n\n");
286 read_counter(counter);
287 428
288 for (counter = 0; counter < nr_counters; counter++) 429 for (counter = 0; counter < nr_counters; counter++)
289 print_counter(counter); 430 print_counter(counter);
290 431
291 432
292 fprintf(stderr, "\n"); 433 fprintf(stderr, "\n");
293 fprintf(stderr, " Wall-clock time elapsed: %12.6f msecs\n", 434 fprintf(stderr, " %14.9f seconds time elapsed.\n",
294 (double)(t1-t0)/1e6); 435 (double)walltime_nsecs_avg/1e9);
295 fprintf(stderr, "\n"); 436 fprintf(stderr, "\n");
296
297 return 0;
298} 437}
299 438
300static volatile int signr = -1; 439static volatile int signr = -1;
@@ -332,11 +471,15 @@ static const struct option options[] = {
332 "scale/normalize counters"), 471 "scale/normalize counters"),
333 OPT_BOOLEAN('v', "verbose", &verbose, 472 OPT_BOOLEAN('v', "verbose", &verbose,
334 "be more verbose (show counter open errors, etc)"), 473 "be more verbose (show counter open errors, etc)"),
474 OPT_INTEGER('r', "repeat", &run_count,
475 "repeat command and print average + stddev (max: 100)"),
335 OPT_END() 476 OPT_END()
336}; 477};
337 478
338int cmd_stat(int argc, const char **argv, const char *prefix) 479int cmd_stat(int argc, const char **argv, const char *prefix)
339{ 480{
481 int status;
482
340 page_size = sysconf(_SC_PAGE_SIZE); 483 page_size = sysconf(_SC_PAGE_SIZE);
341 484
342 memcpy(attrs, default_attrs, sizeof(attrs)); 485 memcpy(attrs, default_attrs, sizeof(attrs));
@@ -344,6 +487,8 @@ int cmd_stat(int argc, const char **argv, const char *prefix)
344 argc = parse_options(argc, argv, options, stat_usage, 0); 487 argc = parse_options(argc, argv, options, stat_usage, 0);
345 if (!argc) 488 if (!argc)
346 usage_with_options(stat_usage, options); 489 usage_with_options(stat_usage, options);
490 if (run_count <= 0 || run_count > MAX_RUN)
491 usage_with_options(stat_usage, options);
347 492
348 if (!nr_counters) 493 if (!nr_counters)
349 nr_counters = 8; 494 nr_counters = 8;
@@ -363,5 +508,14 @@ int cmd_stat(int argc, const char **argv, const char *prefix)
363 signal(SIGALRM, skip_signal); 508 signal(SIGALRM, skip_signal);
364 signal(SIGABRT, skip_signal); 509 signal(SIGABRT, skip_signal);
365 510
366 return do_perf_stat(argc, argv); 511 status = 0;
512 for (run_idx = 0; run_idx < run_count; run_idx++) {
513 if (run_count != 1 && verbose)
514 fprintf(stderr, "[ perf stat: executing run #%d ... ]\n", run_idx+1);
515 status = run_perf_stat(argc, argv);
516 }
517
518 print_stat(argc, argv);
519
520 return status;
367} 521}
diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
index fe338d3c5d7e..5352b5e352ed 100644
--- a/tools/perf/builtin-top.c
+++ b/tools/perf/builtin-top.c
@@ -54,7 +54,7 @@ static int system_wide = 0;
54 54
55static int default_interval = 100000; 55static int default_interval = 100000;
56 56
57static __u64 count_filter = 5; 57static u64 count_filter = 5;
58static int print_entries = 15; 58static int print_entries = 15;
59 59
60static int target_pid = -1; 60static int target_pid = -1;
@@ -79,8 +79,8 @@ static int dump_symtab;
79 * Symbols 79 * Symbols
80 */ 80 */
81 81
82static __u64 min_ip; 82static u64 min_ip;
83static __u64 max_ip = -1ll; 83static u64 max_ip = -1ll;
84 84
85struct sym_entry { 85struct sym_entry {
86 struct rb_node rb_node; 86 struct rb_node rb_node;
@@ -194,7 +194,7 @@ static void print_sym_table(void)
194 100.0 - (100.0*((samples_per_sec-ksamples_per_sec)/samples_per_sec))); 194 100.0 - (100.0*((samples_per_sec-ksamples_per_sec)/samples_per_sec)));
195 195
196 if (nr_counters == 1) { 196 if (nr_counters == 1) {
197 printf("%Ld", attrs[0].sample_period); 197 printf("%Ld", (u64)attrs[0].sample_period);
198 if (freq) 198 if (freq)
199 printf("Hz "); 199 printf("Hz ");
200 else 200 else
@@ -372,7 +372,7 @@ out_delete_dso:
372/* 372/*
373 * Binary search in the histogram table and record the hit: 373 * Binary search in the histogram table and record the hit:
374 */ 374 */
375static void record_ip(__u64 ip, int counter) 375static void record_ip(u64 ip, int counter)
376{ 376{
377 struct symbol *sym = dso__find_symbol(kernel_dso, ip); 377 struct symbol *sym = dso__find_symbol(kernel_dso, ip);
378 378
@@ -392,7 +392,7 @@ static void record_ip(__u64 ip, int counter)
392 samples--; 392 samples--;
393} 393}
394 394
395static void process_event(__u64 ip, int counter) 395static void process_event(u64 ip, int counter)
396{ 396{
397 samples++; 397 samples++;
398 398
@@ -463,15 +463,15 @@ static void mmap_read_counter(struct mmap_data *md)
463 for (; old != head;) { 463 for (; old != head;) {
464 struct ip_event { 464 struct ip_event {
465 struct perf_event_header header; 465 struct perf_event_header header;
466 __u64 ip; 466 u64 ip;
467 __u32 pid, target_pid; 467 u32 pid, target_pid;
468 }; 468 };
469 struct mmap_event { 469 struct mmap_event {
470 struct perf_event_header header; 470 struct perf_event_header header;
471 __u32 pid, target_pid; 471 u32 pid, target_pid;
472 __u64 start; 472 u64 start;
473 __u64 len; 473 u64 len;
474 __u64 pgoff; 474 u64 pgoff;
475 char filename[PATH_MAX]; 475 char filename[PATH_MAX];
476 }; 476 };
477 477
diff --git a/tools/perf/perf.h b/tools/perf/perf.h
index 87a1aca4a424..bccb529dac08 100644
--- a/tools/perf/perf.h
+++ b/tools/perf/perf.h
@@ -19,6 +19,7 @@
19#include <sys/syscall.h> 19#include <sys/syscall.h>
20 20
21#include "../../include/linux/perf_counter.h" 21#include "../../include/linux/perf_counter.h"
22#include "types.h"
22 23
23/* 24/*
24 * prctl(PR_TASK_PERF_COUNTERS_DISABLE) will (cheaply) disable all 25 * prctl(PR_TASK_PERF_COUNTERS_DISABLE) will (cheaply) disable all
@@ -65,4 +66,10 @@ sys_perf_counter_open(struct perf_counter_attr *attr,
65#define MAX_COUNTERS 256 66#define MAX_COUNTERS 256
66#define MAX_NR_CPUS 256 67#define MAX_NR_CPUS 256
67 68
69struct perf_file_header {
70 u64 version;
71 u64 sample_type;
72 u64 data_size;
73};
74
68#endif 75#endif
diff --git a/tools/perf/types.h b/tools/perf/types.h
new file mode 100644
index 000000000000..5e75f9005940
--- /dev/null
+++ b/tools/perf/types.h
@@ -0,0 +1,17 @@
1#ifndef _PERF_TYPES_H
2#define _PERF_TYPES_H
3
4/*
5 * We define u64 as unsigned long long for every architecture
6 * so that we can print it with %Lx without getting warnings.
7 */
8typedef unsigned long long u64;
9typedef signed long long s64;
10typedef unsigned int u32;
11typedef signed int s32;
12typedef unsigned short u16;
13typedef signed short s16;
14typedef unsigned char u8;
15typedef signed char s8;
16
17#endif /* _PERF_TYPES_H */
diff --git a/tools/perf/util/ctype.c b/tools/perf/util/ctype.c
index b90ec004f29c..0b791bd346bc 100644
--- a/tools/perf/util/ctype.c
+++ b/tools/perf/util/ctype.c
@@ -11,16 +11,21 @@ enum {
11 D = GIT_DIGIT, 11 D = GIT_DIGIT,
12 G = GIT_GLOB_SPECIAL, /* *, ?, [, \\ */ 12 G = GIT_GLOB_SPECIAL, /* *, ?, [, \\ */
13 R = GIT_REGEX_SPECIAL, /* $, (, ), +, ., ^, {, | * */ 13 R = GIT_REGEX_SPECIAL, /* $, (, ), +, ., ^, {, | * */
14 P = GIT_PRINT_EXTRA, /* printable - alpha - digit - glob - regex */
15
16 PS = GIT_SPACE | GIT_PRINT_EXTRA,
14}; 17};
15 18
16unsigned char sane_ctype[256] = { 19unsigned char sane_ctype[256] = {
20/* 0 1 2 3 4 5 6 7 8 9 A B C D E F */
21
17 0, 0, 0, 0, 0, 0, 0, 0, 0, S, S, 0, 0, S, 0, 0, /* 0.. 15 */ 22 0, 0, 0, 0, 0, 0, 0, 0, 0, S, S, 0, 0, S, 0, 0, /* 0.. 15 */
18 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 16.. 31 */ 23 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 16.. 31 */
19 S, 0, 0, 0, R, 0, 0, 0, R, R, G, R, 0, 0, R, 0, /* 32.. 47 */ 24 PS,P, P, P, R, P, P, P, R, R, G, R, P, P, R, P, /* 32.. 47 */
20 D, D, D, D, D, D, D, D, D, D, 0, 0, 0, 0, 0, G, /* 48.. 63 */ 25 D, D, D, D, D, D, D, D, D, D, P, P, P, P, P, G, /* 48.. 63 */
21 0, A, A, A, A, A, A, A, A, A, A, A, A, A, A, A, /* 64.. 79 */ 26 P, A, A, A, A, A, A, A, A, A, A, A, A, A, A, A, /* 64.. 79 */
22 A, A, A, A, A, A, A, A, A, A, A, G, G, 0, R, 0, /* 80.. 95 */ 27 A, A, A, A, A, A, A, A, A, A, A, G, G, P, R, P, /* 80.. 95 */
23 0, A, A, A, A, A, A, A, A, A, A, A, A, A, A, A, /* 96..111 */ 28 P, A, A, A, A, A, A, A, A, A, A, A, A, A, A, A, /* 96..111 */
24 A, A, A, A, A, A, A, A, A, A, A, R, R, 0, 0, 0, /* 112..127 */ 29 A, A, A, A, A, A, A, A, A, A, A, R, R, P, P, 0, /* 112..127 */
25 /* Nothing in the 128.. range */ 30 /* Nothing in the 128.. range */
26}; 31};
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index 5a72586e1df0..35d04da38d6a 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -13,8 +13,8 @@ int nr_counters;
13struct perf_counter_attr attrs[MAX_COUNTERS]; 13struct perf_counter_attr attrs[MAX_COUNTERS];
14 14
15struct event_symbol { 15struct event_symbol {
16 __u8 type; 16 u8 type;
17 __u64 config; 17 u64 config;
18 char *symbol; 18 char *symbol;
19}; 19};
20 20
@@ -63,8 +63,8 @@ static char *hw_event_names[] = {
63}; 63};
64 64
65static char *sw_event_names[] = { 65static char *sw_event_names[] = {
66 "cpu-clock-ticks", 66 "cpu-clock-msecs",
67 "task-clock-ticks", 67 "task-clock-msecs",
68 "page-faults", 68 "page-faults",
69 "context-switches", 69 "context-switches",
70 "CPU-migrations", 70 "CPU-migrations",
@@ -96,7 +96,7 @@ static char *hw_cache_result [][MAX_ALIASES] = {
96 96
97char *event_name(int counter) 97char *event_name(int counter)
98{ 98{
99 __u64 config = attrs[counter].config; 99 u64 config = attrs[counter].config;
100 int type = attrs[counter].type; 100 int type = attrs[counter].type;
101 static char buf[32]; 101 static char buf[32];
102 102
@@ -112,7 +112,7 @@ char *event_name(int counter)
112 return "unknown-hardware"; 112 return "unknown-hardware";
113 113
114 case PERF_TYPE_HW_CACHE: { 114 case PERF_TYPE_HW_CACHE: {
115 __u8 cache_type, cache_op, cache_result; 115 u8 cache_type, cache_op, cache_result;
116 static char name[100]; 116 static char name[100];
117 117
118 cache_type = (config >> 0) & 0xff; 118 cache_type = (config >> 0) & 0xff;
@@ -202,7 +202,7 @@ static int parse_generic_hw_symbols(const char *str, struct perf_counter_attr *a
202 */ 202 */
203static int parse_event_symbols(const char *str, struct perf_counter_attr *attr) 203static int parse_event_symbols(const char *str, struct perf_counter_attr *attr)
204{ 204{
205 __u64 config, id; 205 u64 config, id;
206 int type; 206 int type;
207 unsigned int i; 207 unsigned int i;
208 const char *sep, *pstr; 208 const char *sep, *pstr;
diff --git a/tools/perf/util/string.c b/tools/perf/util/string.c
index ec33c0c7f4e2..c93eca9a7be3 100644
--- a/tools/perf/util/string.c
+++ b/tools/perf/util/string.c
@@ -15,7 +15,7 @@ static int hex(char ch)
15 * While we find nice hex chars, build a long_val. 15 * While we find nice hex chars, build a long_val.
16 * Return number of chars processed. 16 * Return number of chars processed.
17 */ 17 */
18int hex2u64(const char *ptr, __u64 *long_val) 18int hex2u64(const char *ptr, u64 *long_val)
19{ 19{
20 const char *p = ptr; 20 const char *p = ptr;
21 *long_val = 0; 21 *long_val = 0;
diff --git a/tools/perf/util/string.h b/tools/perf/util/string.h
index 72812c1c9a7a..37b03255b425 100644
--- a/tools/perf/util/string.h
+++ b/tools/perf/util/string.h
@@ -1,8 +1,8 @@
1#ifndef _PERF_STRING_H_ 1#ifndef _PERF_STRING_H_
2#define _PERF_STRING_H_ 2#define _PERF_STRING_H_
3 3
4#include <linux/types.h> 4#include "../types.h"
5 5
6int hex2u64(const char *ptr, __u64 *val); 6int hex2u64(const char *ptr, u64 *val);
7 7
8#endif 8#endif
diff --git a/tools/perf/util/symbol.c b/tools/perf/util/symbol.c
index 49a55f813712..86e14375e74e 100644
--- a/tools/perf/util/symbol.c
+++ b/tools/perf/util/symbol.c
@@ -9,9 +9,9 @@
9 9
10const char *sym_hist_filter; 10const char *sym_hist_filter;
11 11
12static struct symbol *symbol__new(__u64 start, __u64 len, 12static struct symbol *symbol__new(u64 start, u64 len,
13 const char *name, unsigned int priv_size, 13 const char *name, unsigned int priv_size,
14 __u64 obj_start, int verbose) 14 u64 obj_start, int verbose)
15{ 15{
16 size_t namelen = strlen(name) + 1; 16 size_t namelen = strlen(name) + 1;
17 struct symbol *self = calloc(1, priv_size + sizeof(*self) + namelen); 17 struct symbol *self = calloc(1, priv_size + sizeof(*self) + namelen);
@@ -21,14 +21,14 @@ static struct symbol *symbol__new(__u64 start, __u64 len,
21 21
22 if (verbose >= 2) 22 if (verbose >= 2)
23 printf("new symbol: %016Lx [%08lx]: %s, hist: %p, obj_start: %p\n", 23 printf("new symbol: %016Lx [%08lx]: %s, hist: %p, obj_start: %p\n",
24 (__u64)start, (unsigned long)len, name, self->hist, (void *)(unsigned long)obj_start); 24 (u64)start, (unsigned long)len, name, self->hist, (void *)(unsigned long)obj_start);
25 25
26 self->obj_start= obj_start; 26 self->obj_start= obj_start;
27 self->hist = NULL; 27 self->hist = NULL;
28 self->hist_sum = 0; 28 self->hist_sum = 0;
29 29
30 if (sym_hist_filter && !strcmp(name, sym_hist_filter)) 30 if (sym_hist_filter && !strcmp(name, sym_hist_filter))
31 self->hist = calloc(sizeof(__u64), len); 31 self->hist = calloc(sizeof(u64), len);
32 32
33 if (priv_size) { 33 if (priv_size) {
34 memset(self, 0, priv_size); 34 memset(self, 0, priv_size);
@@ -89,7 +89,7 @@ static void dso__insert_symbol(struct dso *self, struct symbol *sym)
89{ 89{
90 struct rb_node **p = &self->syms.rb_node; 90 struct rb_node **p = &self->syms.rb_node;
91 struct rb_node *parent = NULL; 91 struct rb_node *parent = NULL;
92 const __u64 ip = sym->start; 92 const u64 ip = sym->start;
93 struct symbol *s; 93 struct symbol *s;
94 94
95 while (*p != NULL) { 95 while (*p != NULL) {
@@ -104,7 +104,7 @@ static void dso__insert_symbol(struct dso *self, struct symbol *sym)
104 rb_insert_color(&sym->rb_node, &self->syms); 104 rb_insert_color(&sym->rb_node, &self->syms);
105} 105}
106 106
107struct symbol *dso__find_symbol(struct dso *self, __u64 ip) 107struct symbol *dso__find_symbol(struct dso *self, u64 ip)
108{ 108{
109 struct rb_node *n; 109 struct rb_node *n;
110 110
@@ -151,7 +151,7 @@ static int dso__load_kallsyms(struct dso *self, symbol_filter_t filter, int verb
151 goto out_failure; 151 goto out_failure;
152 152
153 while (!feof(file)) { 153 while (!feof(file)) {
154 __u64 start; 154 u64 start;
155 struct symbol *sym; 155 struct symbol *sym;
156 int line_len, len; 156 int line_len, len;
157 char symbol_type; 157 char symbol_type;
@@ -232,7 +232,7 @@ static int dso__load_perf_map(struct dso *self, symbol_filter_t filter, int verb
232 goto out_failure; 232 goto out_failure;
233 233
234 while (!feof(file)) { 234 while (!feof(file)) {
235 __u64 start, size; 235 u64 start, size;
236 struct symbol *sym; 236 struct symbol *sym;
237 int line_len, len; 237 int line_len, len;
238 238
@@ -353,7 +353,7 @@ static int dso__synthesize_plt_symbols(struct dso *self, Elf *elf,
353{ 353{
354 uint32_t nr_rel_entries, idx; 354 uint32_t nr_rel_entries, idx;
355 GElf_Sym sym; 355 GElf_Sym sym;
356 __u64 plt_offset; 356 u64 plt_offset;
357 GElf_Shdr shdr_plt; 357 GElf_Shdr shdr_plt;
358 struct symbol *f; 358 struct symbol *f;
359 GElf_Shdr shdr_rel_plt; 359 GElf_Shdr shdr_rel_plt;
@@ -523,7 +523,7 @@ static int dso__load_sym(struct dso *self, int fd, const char *name,
523 523
524 elf_symtab__for_each_symbol(syms, nr_syms, index, sym) { 524 elf_symtab__for_each_symbol(syms, nr_syms, index, sym) {
525 struct symbol *f; 525 struct symbol *f;
526 __u64 obj_start; 526 u64 obj_start;
527 527
528 if (!elf_sym__is_function(&sym)) 528 if (!elf_sym__is_function(&sym))
529 continue; 529 continue;
diff --git a/tools/perf/util/symbol.h b/tools/perf/util/symbol.h
index 0d1292bd8270..ea332e56e458 100644
--- a/tools/perf/util/symbol.h
+++ b/tools/perf/util/symbol.h
@@ -2,16 +2,18 @@
2#define _PERF_SYMBOL_ 1 2#define _PERF_SYMBOL_ 1
3 3
4#include <linux/types.h> 4#include <linux/types.h>
5#include "../types.h"
5#include "list.h" 6#include "list.h"
6#include "rbtree.h" 7#include "rbtree.h"
7 8
8struct symbol { 9struct symbol {
9 struct rb_node rb_node; 10 struct rb_node rb_node;
10 __u64 start; 11 u64 start;
11 __u64 end; 12 u64 end;
12 __u64 obj_start; 13 u64 obj_start;
13 __u64 hist_sum; 14 u64 hist_sum;
14 __u64 *hist; 15 u64 *hist;
16 void *priv;
15 char name[0]; 17 char name[0];
16}; 18};
17 19
@@ -19,7 +21,7 @@ struct dso {
19 struct list_head node; 21 struct list_head node;
20 struct rb_root syms; 22 struct rb_root syms;
21 unsigned int sym_priv_size; 23 unsigned int sym_priv_size;
22 struct symbol *(*find_symbol)(struct dso *, __u64 ip); 24 struct symbol *(*find_symbol)(struct dso *, u64 ip);
23 char name[0]; 25 char name[0];
24}; 26};
25 27
@@ -35,7 +37,7 @@ static inline void *dso__sym_priv(struct dso *self, struct symbol *sym)
35 return ((void *)sym) - self->sym_priv_size; 37 return ((void *)sym) - self->sym_priv_size;
36} 38}
37 39
38struct symbol *dso__find_symbol(struct dso *self, __u64 ip); 40struct symbol *dso__find_symbol(struct dso *self, u64 ip);
39 41
40int dso__load_kernel(struct dso *self, const char *vmlinux, 42int dso__load_kernel(struct dso *self, const char *vmlinux,
41 symbol_filter_t filter, int verbose); 43 symbol_filter_t filter, int verbose);
diff --git a/tools/perf/util/util.h b/tools/perf/util/util.h
index 76590a16c271..b8cfed776d81 100644
--- a/tools/perf/util/util.h
+++ b/tools/perf/util/util.h
@@ -100,11 +100,6 @@
100#include <iconv.h> 100#include <iconv.h>
101#endif 101#endif
102 102
103#ifndef NO_OPENSSL
104#include <openssl/ssl.h>
105#include <openssl/err.h>
106#endif
107
108/* On most systems <limits.h> would have given us this, but 103/* On most systems <limits.h> would have given us this, but
109 * not on some systems (e.g. GNU/Hurd). 104 * not on some systems (e.g. GNU/Hurd).
110 */ 105 */
@@ -332,17 +327,20 @@ static inline int has_extension(const char *filename, const char *ext)
332#undef tolower 327#undef tolower
333#undef toupper 328#undef toupper
334extern unsigned char sane_ctype[256]; 329extern unsigned char sane_ctype[256];
335#define GIT_SPACE 0x01 330#define GIT_SPACE 0x01
336#define GIT_DIGIT 0x02 331#define GIT_DIGIT 0x02
337#define GIT_ALPHA 0x04 332#define GIT_ALPHA 0x04
338#define GIT_GLOB_SPECIAL 0x08 333#define GIT_GLOB_SPECIAL 0x08
339#define GIT_REGEX_SPECIAL 0x10 334#define GIT_REGEX_SPECIAL 0x10
335#define GIT_PRINT_EXTRA 0x20
336#define GIT_PRINT 0x3E
340#define sane_istest(x,mask) ((sane_ctype[(unsigned char)(x)] & (mask)) != 0) 337#define sane_istest(x,mask) ((sane_ctype[(unsigned char)(x)] & (mask)) != 0)
341#define isascii(x) (((x) & ~0x7f) == 0) 338#define isascii(x) (((x) & ~0x7f) == 0)
342#define isspace(x) sane_istest(x,GIT_SPACE) 339#define isspace(x) sane_istest(x,GIT_SPACE)
343#define isdigit(x) sane_istest(x,GIT_DIGIT) 340#define isdigit(x) sane_istest(x,GIT_DIGIT)
344#define isalpha(x) sane_istest(x,GIT_ALPHA) 341#define isalpha(x) sane_istest(x,GIT_ALPHA)
345#define isalnum(x) sane_istest(x,GIT_ALPHA | GIT_DIGIT) 342#define isalnum(x) sane_istest(x,GIT_ALPHA | GIT_DIGIT)
343#define isprint(x) sane_istest(x,GIT_PRINT)
346#define is_glob_special(x) sane_istest(x,GIT_GLOB_SPECIAL) 344#define is_glob_special(x) sane_istest(x,GIT_GLOB_SPECIAL)
347#define is_regex_special(x) sane_istest(x,GIT_GLOB_SPECIAL | GIT_REGEX_SPECIAL) 345#define is_regex_special(x) sane_istest(x,GIT_GLOB_SPECIAL | GIT_REGEX_SPECIAL)
348#define tolower(x) sane_case((unsigned char)(x), 0x20) 346#define tolower(x) sane_case((unsigned char)(x), 0x20)
generated by cgit v1.2.2 (git 2.25.0) at 2024-10-04 12:20:01 -0400