aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/oprofile
diff options
context:
space:
mode:
Diffstat (limited to 'arch/x86/oprofile')
-rw-r--r--arch/x86/oprofile/backtrace.c9
-rw-r--r--arch/x86/oprofile/nmi_int.c20
-rw-r--r--arch/x86/oprofile/op_model_amd.c261
-rw-r--r--arch/x86/oprofile/op_model_p4.c6
-rw-r--r--arch/x86/oprofile/op_model_ppro.c21
-rw-r--r--arch/x86/oprofile/op_x86_model.h20
6 files changed, 210 insertions, 127 deletions
diff --git a/arch/x86/oprofile/backtrace.c b/arch/x86/oprofile/backtrace.c
index 044897be021f..3855096c59b8 100644
--- a/arch/x86/oprofile/backtrace.c
+++ b/arch/x86/oprofile/backtrace.c
@@ -41,10 +41,11 @@ static void backtrace_address(void *data, unsigned long addr, int reliable)
41} 41}
42 42
43static struct stacktrace_ops backtrace_ops = { 43static struct stacktrace_ops backtrace_ops = {
44 .warning = backtrace_warning, 44 .warning = backtrace_warning,
45 .warning_symbol = backtrace_warning_symbol, 45 .warning_symbol = backtrace_warning_symbol,
46 .stack = backtrace_stack, 46 .stack = backtrace_stack,
47 .address = backtrace_address, 47 .address = backtrace_address,
48 .walk_stack = print_context_stack,
48}; 49};
49 50
50struct frame_head { 51struct frame_head {
diff --git a/arch/x86/oprofile/nmi_int.c b/arch/x86/oprofile/nmi_int.c
index cb88b1a0bd5f..2c505ee71014 100644
--- a/arch/x86/oprofile/nmi_int.c
+++ b/arch/x86/oprofile/nmi_int.c
@@ -159,7 +159,7 @@ static int nmi_setup_mux(void)
159 159
160 for_each_possible_cpu(i) { 160 for_each_possible_cpu(i) {
161 per_cpu(cpu_msrs, i).multiplex = 161 per_cpu(cpu_msrs, i).multiplex =
162 kmalloc(multiplex_size, GFP_KERNEL); 162 kzalloc(multiplex_size, GFP_KERNEL);
163 if (!per_cpu(cpu_msrs, i).multiplex) 163 if (!per_cpu(cpu_msrs, i).multiplex)
164 return 0; 164 return 0;
165 } 165 }
@@ -179,7 +179,6 @@ static void nmi_cpu_setup_mux(int cpu, struct op_msrs const * const msrs)
179 if (counter_config[i].enabled) { 179 if (counter_config[i].enabled) {
180 multiplex[i].saved = -(u64)counter_config[i].count; 180 multiplex[i].saved = -(u64)counter_config[i].count;
181 } else { 181 } else {
182 multiplex[i].addr = 0;
183 multiplex[i].saved = 0; 182 multiplex[i].saved = 0;
184 } 183 }
185 } 184 }
@@ -189,25 +188,27 @@ static void nmi_cpu_setup_mux(int cpu, struct op_msrs const * const msrs)
189 188
190static void nmi_cpu_save_mpx_registers(struct op_msrs *msrs) 189static void nmi_cpu_save_mpx_registers(struct op_msrs *msrs)
191{ 190{
191 struct op_msr *counters = msrs->counters;
192 struct op_msr *multiplex = msrs->multiplex; 192 struct op_msr *multiplex = msrs->multiplex;
193 int i; 193 int i;
194 194
195 for (i = 0; i < model->num_counters; ++i) { 195 for (i = 0; i < model->num_counters; ++i) {
196 int virt = op_x86_phys_to_virt(i); 196 int virt = op_x86_phys_to_virt(i);
197 if (multiplex[virt].addr) 197 if (counters[i].addr)
198 rdmsrl(multiplex[virt].addr, multiplex[virt].saved); 198 rdmsrl(counters[i].addr, multiplex[virt].saved);
199 } 199 }
200} 200}
201 201
202static void nmi_cpu_restore_mpx_registers(struct op_msrs *msrs) 202static void nmi_cpu_restore_mpx_registers(struct op_msrs *msrs)
203{ 203{
204 struct op_msr *counters = msrs->counters;
204 struct op_msr *multiplex = msrs->multiplex; 205 struct op_msr *multiplex = msrs->multiplex;
205 int i; 206 int i;
206 207
207 for (i = 0; i < model->num_counters; ++i) { 208 for (i = 0; i < model->num_counters; ++i) {
208 int virt = op_x86_phys_to_virt(i); 209 int virt = op_x86_phys_to_virt(i);
209 if (multiplex[virt].addr) 210 if (counters[i].addr)
210 wrmsrl(multiplex[virt].addr, multiplex[virt].saved); 211 wrmsrl(counters[i].addr, multiplex[virt].saved);
211 } 212 }
212} 213}
213 214
@@ -222,7 +223,7 @@ static void nmi_cpu_switch(void *dummy)
222 223
223 /* move to next set */ 224 /* move to next set */
224 si += model->num_counters; 225 si += model->num_counters;
225 if ((si > model->num_virt_counters) || (counter_config[si].count == 0)) 226 if ((si >= model->num_virt_counters) || (counter_config[si].count == 0))
226 per_cpu(switch_index, cpu) = 0; 227 per_cpu(switch_index, cpu) = 0;
227 else 228 else
228 per_cpu(switch_index, cpu) = si; 229 per_cpu(switch_index, cpu) = si;
@@ -303,11 +304,11 @@ static int allocate_msrs(void)
303 304
304 int i; 305 int i;
305 for_each_possible_cpu(i) { 306 for_each_possible_cpu(i) {
306 per_cpu(cpu_msrs, i).counters = kmalloc(counters_size, 307 per_cpu(cpu_msrs, i).counters = kzalloc(counters_size,
307 GFP_KERNEL); 308 GFP_KERNEL);
308 if (!per_cpu(cpu_msrs, i).counters) 309 if (!per_cpu(cpu_msrs, i).counters)
309 return 0; 310 return 0;
310 per_cpu(cpu_msrs, i).controls = kmalloc(controls_size, 311 per_cpu(cpu_msrs, i).controls = kzalloc(controls_size,
311 GFP_KERNEL); 312 GFP_KERNEL);
312 if (!per_cpu(cpu_msrs, i).controls) 313 if (!per_cpu(cpu_msrs, i).controls)
313 return 0; 314 return 0;
@@ -598,6 +599,7 @@ static int __init ppro_init(char **cpu_type)
598 case 15: case 23: 599 case 15: case 23:
599 *cpu_type = "i386/core_2"; 600 *cpu_type = "i386/core_2";
600 break; 601 break;
602 case 0x2e:
601 case 26: 603 case 26:
602 spec = &op_arch_perfmon_spec; 604 spec = &op_arch_perfmon_spec;
603 *cpu_type = "i386/core_i7"; 605 *cpu_type = "i386/core_i7";
diff --git a/arch/x86/oprofile/op_model_amd.c b/arch/x86/oprofile/op_model_amd.c
index 39686c29f03a..090cbbec7dbd 100644
--- a/arch/x86/oprofile/op_model_amd.c
+++ b/arch/x86/oprofile/op_model_amd.c
@@ -22,6 +22,9 @@
22#include <asm/ptrace.h> 22#include <asm/ptrace.h>
23#include <asm/msr.h> 23#include <asm/msr.h>
24#include <asm/nmi.h> 24#include <asm/nmi.h>
25#include <asm/apic.h>
26#include <asm/processor.h>
27#include <asm/cpufeature.h>
25 28
26#include "op_x86_model.h" 29#include "op_x86_model.h"
27#include "op_counter.h" 30#include "op_counter.h"
@@ -43,23 +46,10 @@
43 46
44static unsigned long reset_value[NUM_VIRT_COUNTERS]; 47static unsigned long reset_value[NUM_VIRT_COUNTERS];
45 48
46#ifdef CONFIG_OPROFILE_IBS
47
48/* IbsFetchCtl bits/masks */
49#define IBS_FETCH_RAND_EN (1ULL<<57)
50#define IBS_FETCH_VAL (1ULL<<49)
51#define IBS_FETCH_ENABLE (1ULL<<48)
52#define IBS_FETCH_CNT_MASK 0xFFFF0000ULL
53
54/*IbsOpCtl bits */
55#define IBS_OP_CNT_CTL (1ULL<<19)
56#define IBS_OP_VAL (1ULL<<18)
57#define IBS_OP_ENABLE (1ULL<<17)
58
59#define IBS_FETCH_SIZE 6 49#define IBS_FETCH_SIZE 6
60#define IBS_OP_SIZE 12 50#define IBS_OP_SIZE 12
61 51
62static int has_ibs; /* AMD Family10h and later */ 52static u32 ibs_caps;
63 53
64struct op_ibs_config { 54struct op_ibs_config {
65 unsigned long op_enabled; 55 unsigned long op_enabled;
@@ -71,24 +61,52 @@ struct op_ibs_config {
71}; 61};
72 62
73static struct op_ibs_config ibs_config; 63static struct op_ibs_config ibs_config;
64static u64 ibs_op_ctl;
74 65
75#endif 66/*
67 * IBS cpuid feature detection
68 */
76 69
77#ifdef CONFIG_OPROFILE_EVENT_MULTIPLEX 70#define IBS_CPUID_FEATURES 0x8000001b
71
72/*
73 * Same bit mask as for IBS cpuid feature flags (Fn8000_001B_EAX), but
74 * bit 0 is used to indicate the existence of IBS.
75 */
76#define IBS_CAPS_AVAIL (1LL<<0)
77#define IBS_CAPS_RDWROPCNT (1LL<<3)
78#define IBS_CAPS_OPCNT (1LL<<4)
78 79
79static void op_mux_fill_in_addresses(struct op_msrs * const msrs) 80/*
81 * IBS randomization macros
82 */
83#define IBS_RANDOM_BITS 12
84#define IBS_RANDOM_MASK ((1ULL << IBS_RANDOM_BITS) - 1)
85#define IBS_RANDOM_MAXCNT_OFFSET (1ULL << (IBS_RANDOM_BITS - 5))
86
87static u32 get_ibs_caps(void)
80{ 88{
81 int i; 89 u32 ibs_caps;
90 unsigned int max_level;
82 91
83 for (i = 0; i < NUM_VIRT_COUNTERS; i++) { 92 if (!boot_cpu_has(X86_FEATURE_IBS))
84 int hw_counter = op_x86_virt_to_phys(i); 93 return 0;
85 if (reserve_perfctr_nmi(MSR_K7_PERFCTR0 + i)) 94
86 msrs->multiplex[i].addr = MSR_K7_PERFCTR0 + hw_counter; 95 /* check IBS cpuid feature flags */
87 else 96 max_level = cpuid_eax(0x80000000);
88 msrs->multiplex[i].addr = 0; 97 if (max_level < IBS_CPUID_FEATURES)
89 } 98 return IBS_CAPS_AVAIL;
99
100 ibs_caps = cpuid_eax(IBS_CPUID_FEATURES);
101 if (!(ibs_caps & IBS_CAPS_AVAIL))
102 /* cpuid flags not valid */
103 return IBS_CAPS_AVAIL;
104
105 return ibs_caps;
90} 106}
91 107
108#ifdef CONFIG_OPROFILE_EVENT_MULTIPLEX
109
92static void op_mux_switch_ctrl(struct op_x86_model_spec const *model, 110static void op_mux_switch_ctrl(struct op_x86_model_spec const *model,
93 struct op_msrs const * const msrs) 111 struct op_msrs const * const msrs)
94{ 112{
@@ -98,7 +116,7 @@ static void op_mux_switch_ctrl(struct op_x86_model_spec const *model,
98 /* enable active counters */ 116 /* enable active counters */
99 for (i = 0; i < NUM_COUNTERS; ++i) { 117 for (i = 0; i < NUM_COUNTERS; ++i) {
100 int virt = op_x86_phys_to_virt(i); 118 int virt = op_x86_phys_to_virt(i);
101 if (!counter_config[virt].enabled) 119 if (!reset_value[virt])
102 continue; 120 continue;
103 rdmsrl(msrs->controls[i].addr, val); 121 rdmsrl(msrs->controls[i].addr, val);
104 val &= model->reserved; 122 val &= model->reserved;
@@ -107,10 +125,6 @@ static void op_mux_switch_ctrl(struct op_x86_model_spec const *model,
107 } 125 }
108} 126}
109 127
110#else
111
112static inline void op_mux_fill_in_addresses(struct op_msrs * const msrs) { }
113
114#endif 128#endif
115 129
116/* functions for op_amd_spec */ 130/* functions for op_amd_spec */
@@ -122,18 +136,12 @@ static void op_amd_fill_in_addresses(struct op_msrs * const msrs)
122 for (i = 0; i < NUM_COUNTERS; i++) { 136 for (i = 0; i < NUM_COUNTERS; i++) {
123 if (reserve_perfctr_nmi(MSR_K7_PERFCTR0 + i)) 137 if (reserve_perfctr_nmi(MSR_K7_PERFCTR0 + i))
124 msrs->counters[i].addr = MSR_K7_PERFCTR0 + i; 138 msrs->counters[i].addr = MSR_K7_PERFCTR0 + i;
125 else
126 msrs->counters[i].addr = 0;
127 } 139 }
128 140
129 for (i = 0; i < NUM_CONTROLS; i++) { 141 for (i = 0; i < NUM_CONTROLS; i++) {
130 if (reserve_evntsel_nmi(MSR_K7_EVNTSEL0 + i)) 142 if (reserve_evntsel_nmi(MSR_K7_EVNTSEL0 + i))
131 msrs->controls[i].addr = MSR_K7_EVNTSEL0 + i; 143 msrs->controls[i].addr = MSR_K7_EVNTSEL0 + i;
132 else
133 msrs->controls[i].addr = 0;
134 } 144 }
135
136 op_mux_fill_in_addresses(msrs);
137} 145}
138 146
139static void op_amd_setup_ctrs(struct op_x86_model_spec const *model, 147static void op_amd_setup_ctrs(struct op_x86_model_spec const *model,
@@ -144,7 +152,8 @@ static void op_amd_setup_ctrs(struct op_x86_model_spec const *model,
144 152
145 /* setup reset_value */ 153 /* setup reset_value */
146 for (i = 0; i < NUM_VIRT_COUNTERS; ++i) { 154 for (i = 0; i < NUM_VIRT_COUNTERS; ++i) {
147 if (counter_config[i].enabled) 155 if (counter_config[i].enabled
156 && msrs->counters[op_x86_virt_to_phys(i)].addr)
148 reset_value[i] = counter_config[i].count; 157 reset_value[i] = counter_config[i].count;
149 else 158 else
150 reset_value[i] = 0; 159 reset_value[i] = 0;
@@ -152,9 +161,18 @@ static void op_amd_setup_ctrs(struct op_x86_model_spec const *model,
152 161
153 /* clear all counters */ 162 /* clear all counters */
154 for (i = 0; i < NUM_CONTROLS; ++i) { 163 for (i = 0; i < NUM_CONTROLS; ++i) {
155 if (unlikely(!msrs->controls[i].addr)) 164 if (unlikely(!msrs->controls[i].addr)) {
165 if (counter_config[i].enabled && !smp_processor_id())
166 /*
167 * counter is reserved, this is on all
168 * cpus, so report only for cpu #0
169 */
170 op_x86_warn_reserved(i);
156 continue; 171 continue;
172 }
157 rdmsrl(msrs->controls[i].addr, val); 173 rdmsrl(msrs->controls[i].addr, val);
174 if (val & ARCH_PERFMON_EVENTSEL_ENABLE)
175 op_x86_warn_in_use(i);
158 val &= model->reserved; 176 val &= model->reserved;
159 wrmsrl(msrs->controls[i].addr, val); 177 wrmsrl(msrs->controls[i].addr, val);
160 } 178 }
@@ -169,9 +187,7 @@ static void op_amd_setup_ctrs(struct op_x86_model_spec const *model,
169 /* enable active counters */ 187 /* enable active counters */
170 for (i = 0; i < NUM_COUNTERS; ++i) { 188 for (i = 0; i < NUM_COUNTERS; ++i) {
171 int virt = op_x86_phys_to_virt(i); 189 int virt = op_x86_phys_to_virt(i);
172 if (!counter_config[virt].enabled) 190 if (!reset_value[virt])
173 continue;
174 if (!msrs->counters[i].addr)
175 continue; 191 continue;
176 192
177 /* setup counter registers */ 193 /* setup counter registers */
@@ -185,7 +201,60 @@ static void op_amd_setup_ctrs(struct op_x86_model_spec const *model,
185 } 201 }
186} 202}
187 203
188#ifdef CONFIG_OPROFILE_IBS 204/*
205 * 16-bit Linear Feedback Shift Register (LFSR)
206 *
207 * 16 14 13 11
208 * Feedback polynomial = X + X + X + X + 1
209 */
210static unsigned int lfsr_random(void)
211{
212 static unsigned int lfsr_value = 0xF00D;
213 unsigned int bit;
214
215 /* Compute next bit to shift in */
216 bit = ((lfsr_value >> 0) ^
217 (lfsr_value >> 2) ^
218 (lfsr_value >> 3) ^
219 (lfsr_value >> 5)) & 0x0001;
220
221 /* Advance to next register value */
222 lfsr_value = (lfsr_value >> 1) | (bit << 15);
223
224 return lfsr_value;
225}
226
227/*
228 * IBS software randomization
229 *
230 * The IBS periodic op counter is randomized in software. The lower 12
231 * bits of the 20 bit counter are randomized. IbsOpCurCnt is
232 * initialized with a 12 bit random value.
233 */
234static inline u64 op_amd_randomize_ibs_op(u64 val)
235{
236 unsigned int random = lfsr_random();
237
238 if (!(ibs_caps & IBS_CAPS_RDWROPCNT))
239 /*
240 * Work around if the hw can not write to IbsOpCurCnt
241 *
242 * Randomize the lower 8 bits of the 16 bit
243 * IbsOpMaxCnt [15:0] value in the range of -128 to
244 * +127 by adding/subtracting an offset to the
245 * maximum count (IbsOpMaxCnt).
246 *
247 * To avoid over or underflows and protect upper bits
248 * starting at bit 16, the initial value for
249 * IbsOpMaxCnt must fit in the range from 0x0081 to
250 * 0xff80.
251 */
252 val += (s8)(random >> 4);
253 else
254 val |= (u64)(random & IBS_RANDOM_MASK) << 32;
255
256 return val;
257}
189 258
190static inline void 259static inline void
191op_amd_handle_ibs(struct pt_regs * const regs, 260op_amd_handle_ibs(struct pt_regs * const regs,
@@ -194,7 +263,7 @@ op_amd_handle_ibs(struct pt_regs * const regs,
194 u64 val, ctl; 263 u64 val, ctl;
195 struct op_entry entry; 264 struct op_entry entry;
196 265
197 if (!has_ibs) 266 if (!ibs_caps)
198 return; 267 return;
199 268
200 if (ibs_config.fetch_enabled) { 269 if (ibs_config.fetch_enabled) {
@@ -210,7 +279,7 @@ op_amd_handle_ibs(struct pt_regs * const regs,
210 oprofile_write_commit(&entry); 279 oprofile_write_commit(&entry);
211 280
212 /* reenable the IRQ */ 281 /* reenable the IRQ */
213 ctl &= ~(IBS_FETCH_VAL | IBS_FETCH_CNT_MASK); 282 ctl &= ~(IBS_FETCH_VAL | IBS_FETCH_CNT);
214 ctl |= IBS_FETCH_ENABLE; 283 ctl |= IBS_FETCH_ENABLE;
215 wrmsrl(MSR_AMD64_IBSFETCHCTL, ctl); 284 wrmsrl(MSR_AMD64_IBSFETCHCTL, ctl);
216 } 285 }
@@ -236,8 +305,7 @@ op_amd_handle_ibs(struct pt_regs * const regs,
236 oprofile_write_commit(&entry); 305 oprofile_write_commit(&entry);
237 306
238 /* reenable the IRQ */ 307 /* reenable the IRQ */
239 ctl &= ~IBS_OP_VAL & 0xFFFFFFFF; 308 ctl = op_amd_randomize_ibs_op(ibs_op_ctl);
240 ctl |= IBS_OP_ENABLE;
241 wrmsrl(MSR_AMD64_IBSOPCTL, ctl); 309 wrmsrl(MSR_AMD64_IBSOPCTL, ctl);
242 } 310 }
243 } 311 }
@@ -246,41 +314,57 @@ op_amd_handle_ibs(struct pt_regs * const regs,
246static inline void op_amd_start_ibs(void) 314static inline void op_amd_start_ibs(void)
247{ 315{
248 u64 val; 316 u64 val;
249 if (has_ibs && ibs_config.fetch_enabled) { 317
250 val = (ibs_config.max_cnt_fetch >> 4) & 0xFFFF; 318 if (!ibs_caps)
319 return;
320
321 if (ibs_config.fetch_enabled) {
322 val = (ibs_config.max_cnt_fetch >> 4) & IBS_FETCH_MAX_CNT;
251 val |= ibs_config.rand_en ? IBS_FETCH_RAND_EN : 0; 323 val |= ibs_config.rand_en ? IBS_FETCH_RAND_EN : 0;
252 val |= IBS_FETCH_ENABLE; 324 val |= IBS_FETCH_ENABLE;
253 wrmsrl(MSR_AMD64_IBSFETCHCTL, val); 325 wrmsrl(MSR_AMD64_IBSFETCHCTL, val);
254 } 326 }
255 327
256 if (has_ibs && ibs_config.op_enabled) { 328 if (ibs_config.op_enabled) {
257 val = (ibs_config.max_cnt_op >> 4) & 0xFFFF; 329 ibs_op_ctl = ibs_config.max_cnt_op >> 4;
258 val |= ibs_config.dispatched_ops ? IBS_OP_CNT_CTL : 0; 330 if (!(ibs_caps & IBS_CAPS_RDWROPCNT)) {
259 val |= IBS_OP_ENABLE; 331 /*
332 * IbsOpCurCnt not supported. See
333 * op_amd_randomize_ibs_op() for details.
334 */
335 ibs_op_ctl = clamp(ibs_op_ctl, 0x0081ULL, 0xFF80ULL);
336 } else {
337 /*
338 * The start value is randomized with a
339 * positive offset, we need to compensate it
340 * with the half of the randomized range. Also
341 * avoid underflows.
342 */
343 ibs_op_ctl = min(ibs_op_ctl + IBS_RANDOM_MAXCNT_OFFSET,
344 IBS_OP_MAX_CNT);
345 }
346 if (ibs_caps & IBS_CAPS_OPCNT && ibs_config.dispatched_ops)
347 ibs_op_ctl |= IBS_OP_CNT_CTL;
348 ibs_op_ctl |= IBS_OP_ENABLE;
349 val = op_amd_randomize_ibs_op(ibs_op_ctl);
260 wrmsrl(MSR_AMD64_IBSOPCTL, val); 350 wrmsrl(MSR_AMD64_IBSOPCTL, val);
261 } 351 }
262} 352}
263 353
264static void op_amd_stop_ibs(void) 354static void op_amd_stop_ibs(void)
265{ 355{
266 if (has_ibs && ibs_config.fetch_enabled) 356 if (!ibs_caps)
357 return;
358
359 if (ibs_config.fetch_enabled)
267 /* clear max count and enable */ 360 /* clear max count and enable */
268 wrmsrl(MSR_AMD64_IBSFETCHCTL, 0); 361 wrmsrl(MSR_AMD64_IBSFETCHCTL, 0);
269 362
270 if (has_ibs && ibs_config.op_enabled) 363 if (ibs_config.op_enabled)
271 /* clear max count and enable */ 364 /* clear max count and enable */
272 wrmsrl(MSR_AMD64_IBSOPCTL, 0); 365 wrmsrl(MSR_AMD64_IBSOPCTL, 0);
273} 366}
274 367
275#else
276
277static inline void op_amd_handle_ibs(struct pt_regs * const regs,
278 struct op_msrs const * const msrs) { }
279static inline void op_amd_start_ibs(void) { }
280static inline void op_amd_stop_ibs(void) { }
281
282#endif
283
284static int op_amd_check_ctrs(struct pt_regs * const regs, 368static int op_amd_check_ctrs(struct pt_regs * const regs,
285 struct op_msrs const * const msrs) 369 struct op_msrs const * const msrs)
286{ 370{
@@ -314,7 +398,7 @@ static void op_amd_start(struct op_msrs const * const msrs)
314 if (!reset_value[op_x86_phys_to_virt(i)]) 398 if (!reset_value[op_x86_phys_to_virt(i)])
315 continue; 399 continue;
316 rdmsrl(msrs->controls[i].addr, val); 400 rdmsrl(msrs->controls[i].addr, val);
317 val |= ARCH_PERFMON_EVENTSEL0_ENABLE; 401 val |= ARCH_PERFMON_EVENTSEL_ENABLE;
318 wrmsrl(msrs->controls[i].addr, val); 402 wrmsrl(msrs->controls[i].addr, val);
319 } 403 }
320 404
@@ -334,7 +418,7 @@ static void op_amd_stop(struct op_msrs const * const msrs)
334 if (!reset_value[op_x86_phys_to_virt(i)]) 418 if (!reset_value[op_x86_phys_to_virt(i)])
335 continue; 419 continue;
336 rdmsrl(msrs->controls[i].addr, val); 420 rdmsrl(msrs->controls[i].addr, val);
337 val &= ~ARCH_PERFMON_EVENTSEL0_ENABLE; 421 val &= ~ARCH_PERFMON_EVENTSEL_ENABLE;
338 wrmsrl(msrs->controls[i].addr, val); 422 wrmsrl(msrs->controls[i].addr, val);
339 } 423 }
340 424
@@ -355,8 +439,6 @@ static void op_amd_shutdown(struct op_msrs const * const msrs)
355 } 439 }
356} 440}
357 441
358#ifdef CONFIG_OPROFILE_IBS
359
360static u8 ibs_eilvt_off; 442static u8 ibs_eilvt_off;
361 443
362static inline void apic_init_ibs_nmi_per_cpu(void *arg) 444static inline void apic_init_ibs_nmi_per_cpu(void *arg)
@@ -405,45 +487,36 @@ static int init_ibs_nmi(void)
405 return 1; 487 return 1;
406 } 488 }
407 489
408#ifdef CONFIG_NUMA
409 /* Sanity check */
410 /* Works only for 64bit with proper numa implementation. */
411 if (nodes != num_possible_nodes()) {
412 printk(KERN_DEBUG "Failed to setup CPU node(s) for IBS, "
413 "found: %d, expected %d",
414 nodes, num_possible_nodes());
415 return 1;
416 }
417#endif
418 return 0; 490 return 0;
419} 491}
420 492
421/* uninitialize the APIC for the IBS interrupts if needed */ 493/* uninitialize the APIC for the IBS interrupts if needed */
422static void clear_ibs_nmi(void) 494static void clear_ibs_nmi(void)
423{ 495{
424 if (has_ibs) 496 if (ibs_caps)
425 on_each_cpu(apic_clear_ibs_nmi_per_cpu, NULL, 1); 497 on_each_cpu(apic_clear_ibs_nmi_per_cpu, NULL, 1);
426} 498}
427 499
428/* initialize the APIC for the IBS interrupts if available */ 500/* initialize the APIC for the IBS interrupts if available */
429static void ibs_init(void) 501static void ibs_init(void)
430{ 502{
431 has_ibs = boot_cpu_has(X86_FEATURE_IBS); 503 ibs_caps = get_ibs_caps();
432 504
433 if (!has_ibs) 505 if (!ibs_caps)
434 return; 506 return;
435 507
436 if (init_ibs_nmi()) { 508 if (init_ibs_nmi()) {
437 has_ibs = 0; 509 ibs_caps = 0;
438 return; 510 return;
439 } 511 }
440 512
441 printk(KERN_INFO "oprofile: AMD IBS detected\n"); 513 printk(KERN_INFO "oprofile: AMD IBS detected (0x%08x)\n",
514 (unsigned)ibs_caps);
442} 515}
443 516
444static void ibs_exit(void) 517static void ibs_exit(void)
445{ 518{
446 if (!has_ibs) 519 if (!ibs_caps)
447 return; 520 return;
448 521
449 clear_ibs_nmi(); 522 clear_ibs_nmi();
@@ -463,7 +536,7 @@ static int setup_ibs_files(struct super_block *sb, struct dentry *root)
463 if (ret) 536 if (ret)
464 return ret; 537 return ret;
465 538
466 if (!has_ibs) 539 if (!ibs_caps)
467 return ret; 540 return ret;
468 541
469 /* model specific files */ 542 /* model specific files */
@@ -473,7 +546,7 @@ static int setup_ibs_files(struct super_block *sb, struct dentry *root)
473 ibs_config.fetch_enabled = 0; 546 ibs_config.fetch_enabled = 0;
474 ibs_config.max_cnt_op = 250000; 547 ibs_config.max_cnt_op = 250000;
475 ibs_config.op_enabled = 0; 548 ibs_config.op_enabled = 0;
476 ibs_config.dispatched_ops = 1; 549 ibs_config.dispatched_ops = 0;
477 550
478 dir = oprofilefs_mkdir(sb, root, "ibs_fetch"); 551 dir = oprofilefs_mkdir(sb, root, "ibs_fetch");
479 oprofilefs_create_ulong(sb, dir, "enable", 552 oprofilefs_create_ulong(sb, dir, "enable",
@@ -488,8 +561,9 @@ static int setup_ibs_files(struct super_block *sb, struct dentry *root)
488 &ibs_config.op_enabled); 561 &ibs_config.op_enabled);
489 oprofilefs_create_ulong(sb, dir, "max_count", 562 oprofilefs_create_ulong(sb, dir, "max_count",
490 &ibs_config.max_cnt_op); 563 &ibs_config.max_cnt_op);
491 oprofilefs_create_ulong(sb, dir, "dispatched_ops", 564 if (ibs_caps & IBS_CAPS_OPCNT)
492 &ibs_config.dispatched_ops); 565 oprofilefs_create_ulong(sb, dir, "dispatched_ops",
566 &ibs_config.dispatched_ops);
493 567
494 return 0; 568 return 0;
495} 569}
@@ -507,19 +581,6 @@ static void op_amd_exit(void)
507 ibs_exit(); 581 ibs_exit();
508} 582}
509 583
510#else
511
512/* no IBS support */
513
514static int op_amd_init(struct oprofile_operations *ops)
515{
516 return 0;
517}
518
519static void op_amd_exit(void) {}
520
521#endif /* CONFIG_OPROFILE_IBS */
522
523struct op_x86_model_spec op_amd_spec = { 584struct op_x86_model_spec op_amd_spec = {
524 .num_counters = NUM_COUNTERS, 585 .num_counters = NUM_COUNTERS,
525 .num_controls = NUM_CONTROLS, 586 .num_controls = NUM_CONTROLS,
diff --git a/arch/x86/oprofile/op_model_p4.c b/arch/x86/oprofile/op_model_p4.c
index ac6b354becdf..e6a160a4684a 100644
--- a/arch/x86/oprofile/op_model_p4.c
+++ b/arch/x86/oprofile/op_model_p4.c
@@ -394,12 +394,6 @@ static void p4_fill_in_addresses(struct op_msrs * const msrs)
394 setup_num_counters(); 394 setup_num_counters();
395 stag = get_stagger(); 395 stag = get_stagger();
396 396
397 /* initialize some registers */
398 for (i = 0; i < num_counters; ++i)
399 msrs->counters[i].addr = 0;
400 for (i = 0; i < num_controls; ++i)
401 msrs->controls[i].addr = 0;
402
403 /* the counter & cccr registers we pay attention to */ 397 /* the counter & cccr registers we pay attention to */
404 for (i = 0; i < num_counters; ++i) { 398 for (i = 0; i < num_counters; ++i) {
405 addr = p4_counters[VIRT_CTR(stag, i)].counter_address; 399 addr = p4_counters[VIRT_CTR(stag, i)].counter_address;
diff --git a/arch/x86/oprofile/op_model_ppro.c b/arch/x86/oprofile/op_model_ppro.c
index 8eb05878554c..2bf90fafa7b5 100644
--- a/arch/x86/oprofile/op_model_ppro.c
+++ b/arch/x86/oprofile/op_model_ppro.c
@@ -37,15 +37,11 @@ static void ppro_fill_in_addresses(struct op_msrs * const msrs)
37 for (i = 0; i < num_counters; i++) { 37 for (i = 0; i < num_counters; i++) {
38 if (reserve_perfctr_nmi(MSR_P6_PERFCTR0 + i)) 38 if (reserve_perfctr_nmi(MSR_P6_PERFCTR0 + i))
39 msrs->counters[i].addr = MSR_P6_PERFCTR0 + i; 39 msrs->counters[i].addr = MSR_P6_PERFCTR0 + i;
40 else
41 msrs->counters[i].addr = 0;
42 } 40 }
43 41
44 for (i = 0; i < num_counters; i++) { 42 for (i = 0; i < num_counters; i++) {
45 if (reserve_evntsel_nmi(MSR_P6_EVNTSEL0 + i)) 43 if (reserve_evntsel_nmi(MSR_P6_EVNTSEL0 + i))
46 msrs->controls[i].addr = MSR_P6_EVNTSEL0 + i; 44 msrs->controls[i].addr = MSR_P6_EVNTSEL0 + i;
47 else
48 msrs->controls[i].addr = 0;
49 } 45 }
50} 46}
51 47
@@ -57,7 +53,7 @@ static void ppro_setup_ctrs(struct op_x86_model_spec const *model,
57 int i; 53 int i;
58 54
59 if (!reset_value) { 55 if (!reset_value) {
60 reset_value = kmalloc(sizeof(reset_value[0]) * num_counters, 56 reset_value = kzalloc(sizeof(reset_value[0]) * num_counters,
61 GFP_ATOMIC); 57 GFP_ATOMIC);
62 if (!reset_value) 58 if (!reset_value)
63 return; 59 return;
@@ -82,9 +78,18 @@ static void ppro_setup_ctrs(struct op_x86_model_spec const *model,
82 78
83 /* clear all counters */ 79 /* clear all counters */
84 for (i = 0; i < num_counters; ++i) { 80 for (i = 0; i < num_counters; ++i) {
85 if (unlikely(!msrs->controls[i].addr)) 81 if (unlikely(!msrs->controls[i].addr)) {
82 if (counter_config[i].enabled && !smp_processor_id())
83 /*
84 * counter is reserved, this is on all
85 * cpus, so report only for cpu #0
86 */
87 op_x86_warn_reserved(i);
86 continue; 88 continue;
89 }
87 rdmsrl(msrs->controls[i].addr, val); 90 rdmsrl(msrs->controls[i].addr, val);
91 if (val & ARCH_PERFMON_EVENTSEL_ENABLE)
92 op_x86_warn_in_use(i);
88 val &= model->reserved; 93 val &= model->reserved;
89 wrmsrl(msrs->controls[i].addr, val); 94 wrmsrl(msrs->controls[i].addr, val);
90 } 95 }
@@ -161,7 +166,7 @@ static void ppro_start(struct op_msrs const * const msrs)
161 for (i = 0; i < num_counters; ++i) { 166 for (i = 0; i < num_counters; ++i) {
162 if (reset_value[i]) { 167 if (reset_value[i]) {
163 rdmsrl(msrs->controls[i].addr, val); 168 rdmsrl(msrs->controls[i].addr, val);
164 val |= ARCH_PERFMON_EVENTSEL0_ENABLE; 169 val |= ARCH_PERFMON_EVENTSEL_ENABLE;
165 wrmsrl(msrs->controls[i].addr, val); 170 wrmsrl(msrs->controls[i].addr, val);
166 } 171 }
167 } 172 }
@@ -179,7 +184,7 @@ static void ppro_stop(struct op_msrs const * const msrs)
179 if (!reset_value[i]) 184 if (!reset_value[i])
180 continue; 185 continue;
181 rdmsrl(msrs->controls[i].addr, val); 186 rdmsrl(msrs->controls[i].addr, val);
182 val &= ~ARCH_PERFMON_EVENTSEL0_ENABLE; 187 val &= ~ARCH_PERFMON_EVENTSEL_ENABLE;
183 wrmsrl(msrs->controls[i].addr, val); 188 wrmsrl(msrs->controls[i].addr, val);
184 } 189 }
185} 190}
diff --git a/arch/x86/oprofile/op_x86_model.h b/arch/x86/oprofile/op_x86_model.h
index 7b8e75d16081..ff82a755edd4 100644
--- a/arch/x86/oprofile/op_x86_model.h
+++ b/arch/x86/oprofile/op_x86_model.h
@@ -57,6 +57,26 @@ struct op_x86_model_spec {
57 57
58struct op_counter_config; 58struct op_counter_config;
59 59
60static inline void op_x86_warn_in_use(int counter)
61{
62 /*
63 * The warning indicates an already running counter. If
64 * oprofile doesn't collect data, then try using a different
65 * performance counter on your platform to monitor the desired
66 * event. Delete counter #%d from the desired event by editing
67 * the /usr/share/oprofile/%s/<cpu>/events file. If the event
68 * cannot be monitored by any other counter, contact your
69 * hardware or BIOS vendor.
70 */
71 pr_warning("oprofile: counter #%d on cpu #%d may already be used\n",
72 counter, smp_processor_id());
73}
74
75static inline void op_x86_warn_reserved(int counter)
76{
77 pr_warning("oprofile: counter #%d is already reserved\n", counter);
78}
79
60extern u64 op_x86_get_ctrl(struct op_x86_model_spec const *model, 80extern u64 op_x86_get_ctrl(struct op_x86_model_spec const *model,
61 struct op_counter_config *counter_config); 81 struct op_counter_config *counter_config);
62extern int op_x86_phys_to_virt(int phys); 82extern int op_x86_phys_to_virt(int phys);