aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/include
diff options
context:
space:
mode:
authorSteven Rostedt <srostedt@redhat.com>2010-05-17 22:26:53 -0400
committerSteven Rostedt <rostedt@goodmis.org>2010-05-18 00:35:23 -0400
commitf0218b3e9974f06014b61be8987159f4a20e011e (patch)
tree29a593c4d71ab18cb0c450a34e79bf6bea66877e /arch/x86/include
parent1eaa4787a774c4896518c81f24e8bccaa2244924 (diff)
parent9d192e118a094087494997ea1c8a2faf39af38c5 (diff)
Merge branch 'perf/core' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip into trace/tip/tracing/core-6
Conflicts: include/trace/ftrace.h kernel/trace/trace_kprobe.c Acked-by: Masami Hiramatsu <mhiramat@redhat.com> Acked-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Diffstat (limited to 'arch/x86/include')
-rw-r--r--arch/x86/include/asm/apic.h13
-rw-r--r--arch/x86/include/asm/ds.h302
-rw-r--r--arch/x86/include/asm/hw_breakpoint.h10
-rw-r--r--arch/x86/include/asm/insn.h2
-rw-r--r--arch/x86/include/asm/io.h1
-rw-r--r--arch/x86/include/asm/kprobes.h2
-rw-r--r--arch/x86/include/asm/msr-index.h15
-rw-r--r--arch/x86/include/asm/perf_event.h76
-rw-r--r--arch/x86/include/asm/perf_event_p4.h794
-rw-r--r--arch/x86/include/asm/processor.h35
-rw-r--r--arch/x86/include/asm/ptrace-abi.h57
-rw-r--r--arch/x86/include/asm/ptrace.h6
-rw-r--r--arch/x86/include/asm/thread_info.h8
13 files changed, 872 insertions, 449 deletions
diff --git a/arch/x86/include/asm/apic.h b/arch/x86/include/asm/apic.h
index b4ac2cdcb64f..1fa03e04ae44 100644
--- a/arch/x86/include/asm/apic.h
+++ b/arch/x86/include/asm/apic.h
@@ -373,6 +373,7 @@ extern atomic_t init_deasserted;
373extern int wakeup_secondary_cpu_via_nmi(int apicid, unsigned long start_eip); 373extern int wakeup_secondary_cpu_via_nmi(int apicid, unsigned long start_eip);
374#endif 374#endif
375 375
376#ifdef CONFIG_X86_LOCAL_APIC
376static inline u32 apic_read(u32 reg) 377static inline u32 apic_read(u32 reg)
377{ 378{
378 return apic->read(reg); 379 return apic->read(reg);
@@ -403,10 +404,19 @@ static inline u32 safe_apic_wait_icr_idle(void)
403 return apic->safe_wait_icr_idle(); 404 return apic->safe_wait_icr_idle();
404} 405}
405 406
407#else /* CONFIG_X86_LOCAL_APIC */
408
409static inline u32 apic_read(u32 reg) { return 0; }
410static inline void apic_write(u32 reg, u32 val) { }
411static inline u64 apic_icr_read(void) { return 0; }
412static inline void apic_icr_write(u32 low, u32 high) { }
413static inline void apic_wait_icr_idle(void) { }
414static inline u32 safe_apic_wait_icr_idle(void) { return 0; }
415
416#endif /* CONFIG_X86_LOCAL_APIC */
406 417
407static inline void ack_APIC_irq(void) 418static inline void ack_APIC_irq(void)
408{ 419{
409#ifdef CONFIG_X86_LOCAL_APIC
410 /* 420 /*
411 * ack_APIC_irq() actually gets compiled as a single instruction 421 * ack_APIC_irq() actually gets compiled as a single instruction
412 * ... yummie. 422 * ... yummie.
@@ -414,7 +424,6 @@ static inline void ack_APIC_irq(void)
414 424
415 /* Docs say use 0 for future compatibility */ 425 /* Docs say use 0 for future compatibility */
416 apic_write(APIC_EOI, 0); 426 apic_write(APIC_EOI, 0);
417#endif
418} 427}
419 428
420static inline unsigned default_get_apic_id(unsigned long x) 429static inline unsigned default_get_apic_id(unsigned long x)
diff --git a/arch/x86/include/asm/ds.h b/arch/x86/include/asm/ds.h
deleted file mode 100644
index 70dac199b093..000000000000
--- a/arch/x86/include/asm/ds.h
+++ /dev/null
@@ -1,302 +0,0 @@
1/*
2 * Debug Store (DS) support
3 *
4 * This provides a low-level interface to the hardware's Debug Store
5 * feature that is used for branch trace store (BTS) and
6 * precise-event based sampling (PEBS).
7 *
8 * It manages:
9 * - DS and BTS hardware configuration
10 * - buffer overflow handling (to be done)
11 * - buffer access
12 *
13 * It does not do:
14 * - security checking (is the caller allowed to trace the task)
15 * - buffer allocation (memory accounting)
16 *
17 *
18 * Copyright (C) 2007-2009 Intel Corporation.
19 * Markus Metzger <markus.t.metzger@intel.com>, 2007-2009
20 */
21
22#ifndef _ASM_X86_DS_H
23#define _ASM_X86_DS_H
24
25
26#include <linux/types.h>
27#include <linux/init.h>
28#include <linux/err.h>
29
30
31#ifdef CONFIG_X86_DS
32
33struct task_struct;
34struct ds_context;
35struct ds_tracer;
36struct bts_tracer;
37struct pebs_tracer;
38
39typedef void (*bts_ovfl_callback_t)(struct bts_tracer *);
40typedef void (*pebs_ovfl_callback_t)(struct pebs_tracer *);
41
42
43/*
44 * A list of features plus corresponding macros to talk about them in
45 * the ds_request function's flags parameter.
46 *
47 * We use the enum to index an array of corresponding control bits;
48 * we use the macro to index a flags bit-vector.
49 */
50enum ds_feature {
51 dsf_bts = 0,
52 dsf_bts_kernel,
53#define BTS_KERNEL (1 << dsf_bts_kernel)
54 /* trace kernel-mode branches */
55
56 dsf_bts_user,
57#define BTS_USER (1 << dsf_bts_user)
58 /* trace user-mode branches */
59
60 dsf_bts_overflow,
61 dsf_bts_max,
62 dsf_pebs = dsf_bts_max,
63
64 dsf_pebs_max,
65 dsf_ctl_max = dsf_pebs_max,
66 dsf_bts_timestamps = dsf_ctl_max,
67#define BTS_TIMESTAMPS (1 << dsf_bts_timestamps)
68 /* add timestamps into BTS trace */
69
70#define BTS_USER_FLAGS (BTS_KERNEL | BTS_USER | BTS_TIMESTAMPS)
71};
72
73
74/*
75 * Request BTS or PEBS
76 *
77 * Due to alignement constraints, the actual buffer may be slightly
78 * smaller than the requested or provided buffer.
79 *
80 * Returns a pointer to a tracer structure on success, or
81 * ERR_PTR(errcode) on failure.
82 *
83 * The interrupt threshold is independent from the overflow callback
84 * to allow users to use their own overflow interrupt handling mechanism.
85 *
86 * The function might sleep.
87 *
88 * task: the task to request recording for
89 * cpu: the cpu to request recording for
90 * base: the base pointer for the (non-pageable) buffer;
91 * size: the size of the provided buffer in bytes
92 * ovfl: pointer to a function to be called on buffer overflow;
93 * NULL if cyclic buffer requested
94 * th: the interrupt threshold in records from the end of the buffer;
95 * -1 if no interrupt threshold is requested.
96 * flags: a bit-mask of the above flags
97 */
98extern struct bts_tracer *ds_request_bts_task(struct task_struct *task,
99 void *base, size_t size,
100 bts_ovfl_callback_t ovfl,
101 size_t th, unsigned int flags);
102extern struct bts_tracer *ds_request_bts_cpu(int cpu, void *base, size_t size,
103 bts_ovfl_callback_t ovfl,
104 size_t th, unsigned int flags);
105extern struct pebs_tracer *ds_request_pebs_task(struct task_struct *task,
106 void *base, size_t size,
107 pebs_ovfl_callback_t ovfl,
108 size_t th, unsigned int flags);
109extern struct pebs_tracer *ds_request_pebs_cpu(int cpu,
110 void *base, size_t size,
111 pebs_ovfl_callback_t ovfl,
112 size_t th, unsigned int flags);
113
114/*
115 * Release BTS or PEBS resources
116 * Suspend and resume BTS or PEBS tracing
117 *
118 * Must be called with irq's enabled.
119 *
120 * tracer: the tracer handle returned from ds_request_~()
121 */
122extern void ds_release_bts(struct bts_tracer *tracer);
123extern void ds_suspend_bts(struct bts_tracer *tracer);
124extern void ds_resume_bts(struct bts_tracer *tracer);
125extern void ds_release_pebs(struct pebs_tracer *tracer);
126extern void ds_suspend_pebs(struct pebs_tracer *tracer);
127extern void ds_resume_pebs(struct pebs_tracer *tracer);
128
129/*
130 * Release BTS or PEBS resources
131 * Suspend and resume BTS or PEBS tracing
132 *
133 * Cpu tracers must call this on the traced cpu.
134 * Task tracers must call ds_release_~_noirq() for themselves.
135 *
136 * May be called with irq's disabled.
137 *
138 * Returns 0 if successful;
139 * -EPERM if the cpu tracer does not trace the current cpu.
140 * -EPERM if the task tracer does not trace itself.
141 *
142 * tracer: the tracer handle returned from ds_request_~()
143 */
144extern int ds_release_bts_noirq(struct bts_tracer *tracer);
145extern int ds_suspend_bts_noirq(struct bts_tracer *tracer);
146extern int ds_resume_bts_noirq(struct bts_tracer *tracer);
147extern int ds_release_pebs_noirq(struct pebs_tracer *tracer);
148extern int ds_suspend_pebs_noirq(struct pebs_tracer *tracer);
149extern int ds_resume_pebs_noirq(struct pebs_tracer *tracer);
150
151
152/*
153 * The raw DS buffer state as it is used for BTS and PEBS recording.
154 *
155 * This is the low-level, arch-dependent interface for working
156 * directly on the raw trace data.
157 */
158struct ds_trace {
159 /* the number of bts/pebs records */
160 size_t n;
161 /* the size of a bts/pebs record in bytes */
162 size_t size;
163 /* pointers into the raw buffer:
164 - to the first entry */
165 void *begin;
166 /* - one beyond the last entry */
167 void *end;
168 /* - one beyond the newest entry */
169 void *top;
170 /* - the interrupt threshold */
171 void *ith;
172 /* flags given on ds_request() */
173 unsigned int flags;
174};
175
176/*
177 * An arch-independent view on branch trace data.
178 */
179enum bts_qualifier {
180 bts_invalid,
181#define BTS_INVALID bts_invalid
182
183 bts_branch,
184#define BTS_BRANCH bts_branch
185
186 bts_task_arrives,
187#define BTS_TASK_ARRIVES bts_task_arrives
188
189 bts_task_departs,
190#define BTS_TASK_DEPARTS bts_task_departs
191
192 bts_qual_bit_size = 4,
193 bts_qual_max = (1 << bts_qual_bit_size),
194};
195
196struct bts_struct {
197 __u64 qualifier;
198 union {
199 /* BTS_BRANCH */
200 struct {
201 __u64 from;
202 __u64 to;
203 } lbr;
204 /* BTS_TASK_ARRIVES or BTS_TASK_DEPARTS */
205 struct {
206 __u64 clock;
207 pid_t pid;
208 } event;
209 } variant;
210};
211
212
213/*
214 * The BTS state.
215 *
216 * This gives access to the raw DS state and adds functions to provide
217 * an arch-independent view of the BTS data.
218 */
219struct bts_trace {
220 struct ds_trace ds;
221
222 int (*read)(struct bts_tracer *tracer, const void *at,
223 struct bts_struct *out);
224 int (*write)(struct bts_tracer *tracer, const struct bts_struct *in);
225};
226
227
228/*
229 * The PEBS state.
230 *
231 * This gives access to the raw DS state and the PEBS-specific counter
232 * reset value.
233 */
234struct pebs_trace {
235 struct ds_trace ds;
236
237 /* the number of valid counters in the below array */
238 unsigned int counters;
239
240#define MAX_PEBS_COUNTERS 4
241 /* the counter reset value */
242 unsigned long long counter_reset[MAX_PEBS_COUNTERS];
243};
244
245
246/*
247 * Read the BTS or PEBS trace.
248 *
249 * Returns a view on the trace collected for the parameter tracer.
250 *
251 * The view remains valid as long as the traced task is not running or
252 * the tracer is suspended.
253 * Writes into the trace buffer are not reflected.
254 *
255 * tracer: the tracer handle returned from ds_request_~()
256 */
257extern const struct bts_trace *ds_read_bts(struct bts_tracer *tracer);
258extern const struct pebs_trace *ds_read_pebs(struct pebs_tracer *tracer);
259
260
261/*
262 * Reset the write pointer of the BTS/PEBS buffer.
263 *
264 * Returns 0 on success; -Eerrno on error
265 *
266 * tracer: the tracer handle returned from ds_request_~()
267 */
268extern int ds_reset_bts(struct bts_tracer *tracer);
269extern int ds_reset_pebs(struct pebs_tracer *tracer);
270
271/*
272 * Set the PEBS counter reset value.
273 *
274 * Returns 0 on success; -Eerrno on error
275 *
276 * tracer: the tracer handle returned from ds_request_pebs()
277 * counter: the index of the counter
278 * value: the new counter reset value
279 */
280extern int ds_set_pebs_reset(struct pebs_tracer *tracer,
281 unsigned int counter, u64 value);
282
283/*
284 * Initialization
285 */
286struct cpuinfo_x86;
287extern void __cpuinit ds_init_intel(struct cpuinfo_x86 *);
288
289/*
290 * Context switch work
291 */
292extern void ds_switch_to(struct task_struct *prev, struct task_struct *next);
293
294#else /* CONFIG_X86_DS */
295
296struct cpuinfo_x86;
297static inline void __cpuinit ds_init_intel(struct cpuinfo_x86 *ignored) {}
298static inline void ds_switch_to(struct task_struct *prev,
299 struct task_struct *next) {}
300
301#endif /* CONFIG_X86_DS */
302#endif /* _ASM_X86_DS_H */
diff --git a/arch/x86/include/asm/hw_breakpoint.h b/arch/x86/include/asm/hw_breakpoint.h
index 2a1bd8f4f23a..942255310e6a 100644
--- a/arch/x86/include/asm/hw_breakpoint.h
+++ b/arch/x86/include/asm/hw_breakpoint.h
@@ -41,12 +41,16 @@ struct arch_hw_breakpoint {
41/* Total number of available HW breakpoint registers */ 41/* Total number of available HW breakpoint registers */
42#define HBP_NUM 4 42#define HBP_NUM 4
43 43
44static inline int hw_breakpoint_slots(int type)
45{
46 return HBP_NUM;
47}
48
44struct perf_event; 49struct perf_event;
45struct pmu; 50struct pmu;
46 51
47extern int arch_check_va_in_userspace(unsigned long va, u8 hbp_len); 52extern int arch_check_bp_in_kernelspace(struct perf_event *bp);
48extern int arch_validate_hwbkpt_settings(struct perf_event *bp, 53extern int arch_validate_hwbkpt_settings(struct perf_event *bp);
49 struct task_struct *tsk);
50extern int hw_breakpoint_exceptions_notify(struct notifier_block *unused, 54extern int hw_breakpoint_exceptions_notify(struct notifier_block *unused,
51 unsigned long val, void *data); 55 unsigned long val, void *data);
52 56
diff --git a/arch/x86/include/asm/insn.h b/arch/x86/include/asm/insn.h
index 96c2e0ad04ca..88c765e16410 100644
--- a/arch/x86/include/asm/insn.h
+++ b/arch/x86/include/asm/insn.h
@@ -68,6 +68,8 @@ struct insn {
68 const insn_byte_t *next_byte; 68 const insn_byte_t *next_byte;
69}; 69};
70 70
71#define MAX_INSN_SIZE 16
72
71#define X86_MODRM_MOD(modrm) (((modrm) & 0xc0) >> 6) 73#define X86_MODRM_MOD(modrm) (((modrm) & 0xc0) >> 6)
72#define X86_MODRM_REG(modrm) (((modrm) & 0x38) >> 3) 74#define X86_MODRM_REG(modrm) (((modrm) & 0x38) >> 3)
73#define X86_MODRM_RM(modrm) ((modrm) & 0x07) 75#define X86_MODRM_RM(modrm) ((modrm) & 0x07)
diff --git a/arch/x86/include/asm/io.h b/arch/x86/include/asm/io.h
index a1dcfa3ab17d..30a3e9776123 100644
--- a/arch/x86/include/asm/io.h
+++ b/arch/x86/include/asm/io.h
@@ -347,6 +347,7 @@ extern void __iomem *early_ioremap(resource_size_t phys_addr,
347extern void __iomem *early_memremap(resource_size_t phys_addr, 347extern void __iomem *early_memremap(resource_size_t phys_addr,
348 unsigned long size); 348 unsigned long size);
349extern void early_iounmap(void __iomem *addr, unsigned long size); 349extern void early_iounmap(void __iomem *addr, unsigned long size);
350extern void fixup_early_ioremap(void);
350 351
351#define IO_SPACE_LIMIT 0xffff 352#define IO_SPACE_LIMIT 0xffff
352 353
diff --git a/arch/x86/include/asm/kprobes.h b/arch/x86/include/asm/kprobes.h
index 4ffa345a8ccb..547882539157 100644
--- a/arch/x86/include/asm/kprobes.h
+++ b/arch/x86/include/asm/kprobes.h
@@ -24,6 +24,7 @@
24#include <linux/types.h> 24#include <linux/types.h>
25#include <linux/ptrace.h> 25#include <linux/ptrace.h>
26#include <linux/percpu.h> 26#include <linux/percpu.h>
27#include <asm/insn.h>
27 28
28#define __ARCH_WANT_KPROBES_INSN_SLOT 29#define __ARCH_WANT_KPROBES_INSN_SLOT
29 30
@@ -36,7 +37,6 @@ typedef u8 kprobe_opcode_t;
36#define RELATIVEJUMP_SIZE 5 37#define RELATIVEJUMP_SIZE 5
37#define RELATIVECALL_OPCODE 0xe8 38#define RELATIVECALL_OPCODE 0xe8
38#define RELATIVE_ADDR_SIZE 4 39#define RELATIVE_ADDR_SIZE 4
39#define MAX_INSN_SIZE 16
40#define MAX_STACK_SIZE 64 40#define MAX_STACK_SIZE 64
41#define MIN_STACK_SIZE(ADDR) \ 41#define MIN_STACK_SIZE(ADDR) \
42 (((MAX_STACK_SIZE) < (((unsigned long)current_thread_info()) + \ 42 (((MAX_STACK_SIZE) < (((unsigned long)current_thread_info()) + \
diff --git a/arch/x86/include/asm/msr-index.h b/arch/x86/include/asm/msr-index.h
index 4604e6a54d36..bc473acfa7f9 100644
--- a/arch/x86/include/asm/msr-index.h
+++ b/arch/x86/include/asm/msr-index.h
@@ -71,11 +71,14 @@
71#define MSR_IA32_LASTINTTOIP 0x000001de 71#define MSR_IA32_LASTINTTOIP 0x000001de
72 72
73/* DEBUGCTLMSR bits (others vary by model): */ 73/* DEBUGCTLMSR bits (others vary by model): */
74#define _DEBUGCTLMSR_LBR 0 /* last branch recording */ 74#define DEBUGCTLMSR_LBR (1UL << 0) /* last branch recording */
75#define _DEBUGCTLMSR_BTF 1 /* single-step on branches */ 75#define DEBUGCTLMSR_BTF (1UL << 1) /* single-step on branches */
76 76#define DEBUGCTLMSR_TR (1UL << 6)
77#define DEBUGCTLMSR_LBR (1UL << _DEBUGCTLMSR_LBR) 77#define DEBUGCTLMSR_BTS (1UL << 7)
78#define DEBUGCTLMSR_BTF (1UL << _DEBUGCTLMSR_BTF) 78#define DEBUGCTLMSR_BTINT (1UL << 8)
79#define DEBUGCTLMSR_BTS_OFF_OS (1UL << 9)
80#define DEBUGCTLMSR_BTS_OFF_USR (1UL << 10)
81#define DEBUGCTLMSR_FREEZE_LBRS_ON_PMI (1UL << 11)
79 82
80#define MSR_IA32_MC0_CTL 0x00000400 83#define MSR_IA32_MC0_CTL 0x00000400
81#define MSR_IA32_MC0_STATUS 0x00000401 84#define MSR_IA32_MC0_STATUS 0x00000401
@@ -359,6 +362,8 @@
359#define MSR_P4_U2L_ESCR0 0x000003b0 362#define MSR_P4_U2L_ESCR0 0x000003b0
360#define MSR_P4_U2L_ESCR1 0x000003b1 363#define MSR_P4_U2L_ESCR1 0x000003b1
361 364
365#define MSR_P4_PEBS_MATRIX_VERT 0x000003f2
366
362/* Intel Core-based CPU performance counters */ 367/* Intel Core-based CPU performance counters */
363#define MSR_CORE_PERF_FIXED_CTR0 0x00000309 368#define MSR_CORE_PERF_FIXED_CTR0 0x00000309
364#define MSR_CORE_PERF_FIXED_CTR1 0x0000030a 369#define MSR_CORE_PERF_FIXED_CTR1 0x0000030a
diff --git a/arch/x86/include/asm/perf_event.h b/arch/x86/include/asm/perf_event.h
index db6109a885a7..254883d0c7e0 100644
--- a/arch/x86/include/asm/perf_event.h
+++ b/arch/x86/include/asm/perf_event.h
@@ -5,7 +5,7 @@
5 * Performance event hw details: 5 * Performance event hw details:
6 */ 6 */
7 7
8#define X86_PMC_MAX_GENERIC 8 8#define X86_PMC_MAX_GENERIC 32
9#define X86_PMC_MAX_FIXED 3 9#define X86_PMC_MAX_FIXED 3
10 10
11#define X86_PMC_IDX_GENERIC 0 11#define X86_PMC_IDX_GENERIC 0
@@ -18,39 +18,31 @@
18#define MSR_ARCH_PERFMON_EVENTSEL0 0x186 18#define MSR_ARCH_PERFMON_EVENTSEL0 0x186
19#define MSR_ARCH_PERFMON_EVENTSEL1 0x187 19#define MSR_ARCH_PERFMON_EVENTSEL1 0x187
20 20
21#define ARCH_PERFMON_EVENTSEL_ENABLE (1 << 22) 21#define ARCH_PERFMON_EVENTSEL_EVENT 0x000000FFULL
22#define ARCH_PERFMON_EVENTSEL_ANY (1 << 21) 22#define ARCH_PERFMON_EVENTSEL_UMASK 0x0000FF00ULL
23#define ARCH_PERFMON_EVENTSEL_INT (1 << 20) 23#define ARCH_PERFMON_EVENTSEL_USR (1ULL << 16)
24#define ARCH_PERFMON_EVENTSEL_OS (1 << 17) 24#define ARCH_PERFMON_EVENTSEL_OS (1ULL << 17)
25#define ARCH_PERFMON_EVENTSEL_USR (1 << 16) 25#define ARCH_PERFMON_EVENTSEL_EDGE (1ULL << 18)
26 26#define ARCH_PERFMON_EVENTSEL_INT (1ULL << 20)
27/* 27#define ARCH_PERFMON_EVENTSEL_ANY (1ULL << 21)
28 * Includes eventsel and unit mask as well: 28#define ARCH_PERFMON_EVENTSEL_ENABLE (1ULL << 22)
29 */ 29#define ARCH_PERFMON_EVENTSEL_INV (1ULL << 23)
30 30#define ARCH_PERFMON_EVENTSEL_CMASK 0xFF000000ULL
31 31
32#define INTEL_ARCH_EVTSEL_MASK 0x000000FFULL 32#define AMD64_EVENTSEL_EVENT \
33#define INTEL_ARCH_UNIT_MASK 0x0000FF00ULL 33 (ARCH_PERFMON_EVENTSEL_EVENT | (0x0FULL << 32))
34#define INTEL_ARCH_EDGE_MASK 0x00040000ULL 34#define INTEL_ARCH_EVENT_MASK \
35#define INTEL_ARCH_INV_MASK 0x00800000ULL 35 (ARCH_PERFMON_EVENTSEL_UMASK | ARCH_PERFMON_EVENTSEL_EVENT)
36#define INTEL_ARCH_CNT_MASK 0xFF000000ULL 36
37#define INTEL_ARCH_EVENT_MASK (INTEL_ARCH_UNIT_MASK|INTEL_ARCH_EVTSEL_MASK) 37#define X86_RAW_EVENT_MASK \
38 38 (ARCH_PERFMON_EVENTSEL_EVENT | \
39/* 39 ARCH_PERFMON_EVENTSEL_UMASK | \
40 * filter mask to validate fixed counter events. 40 ARCH_PERFMON_EVENTSEL_EDGE | \
41 * the following filters disqualify for fixed counters: 41 ARCH_PERFMON_EVENTSEL_INV | \
42 * - inv 42 ARCH_PERFMON_EVENTSEL_CMASK)
43 * - edge 43#define AMD64_RAW_EVENT_MASK \
44 * - cnt-mask 44 (X86_RAW_EVENT_MASK | \
45 * The other filters are supported by fixed counters. 45 AMD64_EVENTSEL_EVENT)
46 * The any-thread option is supported starting with v3.
47 */
48#define INTEL_ARCH_FIXED_MASK \
49 (INTEL_ARCH_CNT_MASK| \
50 INTEL_ARCH_INV_MASK| \
51 INTEL_ARCH_EDGE_MASK|\
52 INTEL_ARCH_UNIT_MASK|\
53 INTEL_ARCH_EVENT_MASK)
54 46
55#define ARCH_PERFMON_UNHALTED_CORE_CYCLES_SEL 0x3c 47#define ARCH_PERFMON_UNHALTED_CORE_CYCLES_SEL 0x3c
56#define ARCH_PERFMON_UNHALTED_CORE_CYCLES_UMASK (0x00 << 8) 48#define ARCH_PERFMON_UNHALTED_CORE_CYCLES_UMASK (0x00 << 8)
@@ -67,7 +59,7 @@
67union cpuid10_eax { 59union cpuid10_eax {
68 struct { 60 struct {
69 unsigned int version_id:8; 61 unsigned int version_id:8;
70 unsigned int num_events:8; 62 unsigned int num_counters:8;
71 unsigned int bit_width:8; 63 unsigned int bit_width:8;
72 unsigned int mask_length:8; 64 unsigned int mask_length:8;
73 } split; 65 } split;
@@ -76,7 +68,7 @@ union cpuid10_eax {
76 68
77union cpuid10_edx { 69union cpuid10_edx {
78 struct { 70 struct {
79 unsigned int num_events_fixed:4; 71 unsigned int num_counters_fixed:4;
80 unsigned int reserved:28; 72 unsigned int reserved:28;
81 } split; 73 } split;
82 unsigned int full; 74 unsigned int full;
@@ -136,6 +128,18 @@ extern void perf_events_lapic_init(void);
136 128
137#define PERF_EVENT_INDEX_OFFSET 0 129#define PERF_EVENT_INDEX_OFFSET 0
138 130
131/*
132 * Abuse bit 3 of the cpu eflags register to indicate proper PEBS IP fixups.
133 * This flag is otherwise unused and ABI specified to be 0, so nobody should
134 * care what we do with it.
135 */
136#define PERF_EFLAGS_EXACT (1UL << 3)
137
138struct pt_regs;
139extern unsigned long perf_instruction_pointer(struct pt_regs *regs);
140extern unsigned long perf_misc_flags(struct pt_regs *regs);
141#define perf_misc_flags(regs) perf_misc_flags(regs)
142
139#else 143#else
140static inline void init_hw_perf_events(void) { } 144static inline void init_hw_perf_events(void) { }
141static inline void perf_events_lapic_init(void) { } 145static inline void perf_events_lapic_init(void) { }
diff --git a/arch/x86/include/asm/perf_event_p4.h b/arch/x86/include/asm/perf_event_p4.h
new file mode 100644
index 000000000000..b05400a542ff
--- /dev/null
+++ b/arch/x86/include/asm/perf_event_p4.h
@@ -0,0 +1,794 @@
1/*
2 * Netburst Perfomance Events (P4, old Xeon)
3 */
4
5#ifndef PERF_EVENT_P4_H
6#define PERF_EVENT_P4_H
7
8#include <linux/cpu.h>
9#include <linux/bitops.h>
10
11/*
12 * NetBurst has perfomance MSRs shared between
13 * threads if HT is turned on, ie for both logical
14 * processors (mem: in turn in Atom with HT support
15 * perf-MSRs are not shared and every thread has its
16 * own perf-MSRs set)
17 */
18#define ARCH_P4_TOTAL_ESCR (46)
19#define ARCH_P4_RESERVED_ESCR (2) /* IQ_ESCR(0,1) not always present */
20#define ARCH_P4_MAX_ESCR (ARCH_P4_TOTAL_ESCR - ARCH_P4_RESERVED_ESCR)
21#define ARCH_P4_MAX_CCCR (18)
22#define ARCH_P4_MAX_COUNTER (ARCH_P4_MAX_CCCR / 2)
23
24#define P4_ESCR_EVENT_MASK 0x7e000000U
25#define P4_ESCR_EVENT_SHIFT 25
26#define P4_ESCR_EVENTMASK_MASK 0x01fffe00U
27#define P4_ESCR_EVENTMASK_SHIFT 9
28#define P4_ESCR_TAG_MASK 0x000001e0U
29#define P4_ESCR_TAG_SHIFT 5
30#define P4_ESCR_TAG_ENABLE 0x00000010U
31#define P4_ESCR_T0_OS 0x00000008U
32#define P4_ESCR_T0_USR 0x00000004U
33#define P4_ESCR_T1_OS 0x00000002U
34#define P4_ESCR_T1_USR 0x00000001U
35
36#define P4_ESCR_EVENT(v) ((v) << P4_ESCR_EVENT_SHIFT)
37#define P4_ESCR_EMASK(v) ((v) << P4_ESCR_EVENTMASK_SHIFT)
38#define P4_ESCR_TAG(v) ((v) << P4_ESCR_TAG_SHIFT)
39
40/* Non HT mask */
41#define P4_ESCR_MASK \
42 (P4_ESCR_EVENT_MASK | \
43 P4_ESCR_EVENTMASK_MASK | \
44 P4_ESCR_TAG_MASK | \
45 P4_ESCR_TAG_ENABLE | \
46 P4_ESCR_T0_OS | \
47 P4_ESCR_T0_USR)
48
49/* HT mask */
50#define P4_ESCR_MASK_HT \
51 (P4_ESCR_MASK | P4_ESCR_T1_OS | P4_ESCR_T1_USR)
52
53#define P4_CCCR_OVF 0x80000000U
54#define P4_CCCR_CASCADE 0x40000000U
55#define P4_CCCR_OVF_PMI_T0 0x04000000U
56#define P4_CCCR_OVF_PMI_T1 0x08000000U
57#define P4_CCCR_FORCE_OVF 0x02000000U
58#define P4_CCCR_EDGE 0x01000000U
59#define P4_CCCR_THRESHOLD_MASK 0x00f00000U
60#define P4_CCCR_THRESHOLD_SHIFT 20
61#define P4_CCCR_COMPLEMENT 0x00080000U
62#define P4_CCCR_COMPARE 0x00040000U
63#define P4_CCCR_ESCR_SELECT_MASK 0x0000e000U
64#define P4_CCCR_ESCR_SELECT_SHIFT 13
65#define P4_CCCR_ENABLE 0x00001000U
66#define P4_CCCR_THREAD_SINGLE 0x00010000U
67#define P4_CCCR_THREAD_BOTH 0x00020000U
68#define P4_CCCR_THREAD_ANY 0x00030000U
69#define P4_CCCR_RESERVED 0x00000fffU
70
71#define P4_CCCR_THRESHOLD(v) ((v) << P4_CCCR_THRESHOLD_SHIFT)
72#define P4_CCCR_ESEL(v) ((v) << P4_CCCR_ESCR_SELECT_SHIFT)
73
74/* Custom bits in reerved CCCR area */
75#define P4_CCCR_CACHE_OPS_MASK 0x0000003fU
76
77
78/* Non HT mask */
79#define P4_CCCR_MASK \
80 (P4_CCCR_OVF | \
81 P4_CCCR_CASCADE | \
82 P4_CCCR_OVF_PMI_T0 | \
83 P4_CCCR_FORCE_OVF | \
84 P4_CCCR_EDGE | \
85 P4_CCCR_THRESHOLD_MASK | \
86 P4_CCCR_COMPLEMENT | \
87 P4_CCCR_COMPARE | \
88 P4_CCCR_ESCR_SELECT_MASK | \
89 P4_CCCR_ENABLE)
90
91/* HT mask */
92#define P4_CCCR_MASK_HT (P4_CCCR_MASK | P4_CCCR_THREAD_ANY)
93
94#define P4_GEN_ESCR_EMASK(class, name, bit) \
95 class##__##name = ((1 << bit) << P4_ESCR_EVENTMASK_SHIFT)
96#define P4_ESCR_EMASK_BIT(class, name) class##__##name
97
98/*
99 * config field is 64bit width and consists of
100 * HT << 63 | ESCR << 32 | CCCR
101 * where HT is HyperThreading bit (since ESCR
102 * has it reserved we may use it for own purpose)
103 *
104 * note that this is NOT the addresses of respective
105 * ESCR and CCCR but rather an only packed value should
106 * be unpacked and written to a proper addresses
107 *
108 * the base idea is to pack as much info as
109 * possible
110 */
111#define p4_config_pack_escr(v) (((u64)(v)) << 32)
112#define p4_config_pack_cccr(v) (((u64)(v)) & 0xffffffffULL)
113#define p4_config_unpack_escr(v) (((u64)(v)) >> 32)
114#define p4_config_unpack_cccr(v) (((u64)(v)) & 0xffffffffULL)
115
116#define p4_config_unpack_emask(v) \
117 ({ \
118 u32 t = p4_config_unpack_escr((v)); \
119 t = t & P4_ESCR_EVENTMASK_MASK; \
120 t = t >> P4_ESCR_EVENTMASK_SHIFT; \
121 t; \
122 })
123
124#define p4_config_unpack_event(v) \
125 ({ \
126 u32 t = p4_config_unpack_escr((v)); \
127 t = t & P4_ESCR_EVENT_MASK; \
128 t = t >> P4_ESCR_EVENT_SHIFT; \
129 t; \
130 })
131
132#define p4_config_unpack_cache_event(v) (((u64)(v)) & P4_CCCR_CACHE_OPS_MASK)
133
134#define P4_CONFIG_HT_SHIFT 63
135#define P4_CONFIG_HT (1ULL << P4_CONFIG_HT_SHIFT)
136
137static inline bool p4_is_event_cascaded(u64 config)
138{
139 u32 cccr = p4_config_unpack_cccr(config);
140 return !!(cccr & P4_CCCR_CASCADE);
141}
142
143static inline int p4_ht_config_thread(u64 config)
144{
145 return !!(config & P4_CONFIG_HT);
146}
147
148static inline u64 p4_set_ht_bit(u64 config)
149{
150 return config | P4_CONFIG_HT;
151}
152
153static inline u64 p4_clear_ht_bit(u64 config)
154{
155 return config & ~P4_CONFIG_HT;
156}
157
158static inline int p4_ht_active(void)
159{
160#ifdef CONFIG_SMP
161 return smp_num_siblings > 1;
162#endif
163 return 0;
164}
165
166static inline int p4_ht_thread(int cpu)
167{
168#ifdef CONFIG_SMP
169 if (smp_num_siblings == 2)
170 return cpu != cpumask_first(__get_cpu_var(cpu_sibling_map));
171#endif
172 return 0;
173}
174
175static inline int p4_should_swap_ts(u64 config, int cpu)
176{
177 return p4_ht_config_thread(config) ^ p4_ht_thread(cpu);
178}
179
180static inline u32 p4_default_cccr_conf(int cpu)
181{
182 /*
183 * Note that P4_CCCR_THREAD_ANY is "required" on
184 * non-HT machines (on HT machines we count TS events
185 * regardless the state of second logical processor
186 */
187 u32 cccr = P4_CCCR_THREAD_ANY;
188
189 if (!p4_ht_thread(cpu))
190 cccr |= P4_CCCR_OVF_PMI_T0;
191 else
192 cccr |= P4_CCCR_OVF_PMI_T1;
193
194 return cccr;
195}
196
197static inline u32 p4_default_escr_conf(int cpu, int exclude_os, int exclude_usr)
198{
199 u32 escr = 0;
200
201 if (!p4_ht_thread(cpu)) {
202 if (!exclude_os)
203 escr |= P4_ESCR_T0_OS;
204 if (!exclude_usr)
205 escr |= P4_ESCR_T0_USR;
206 } else {
207 if (!exclude_os)
208 escr |= P4_ESCR_T1_OS;
209 if (!exclude_usr)
210 escr |= P4_ESCR_T1_USR;
211 }
212
213 return escr;
214}
215
216enum P4_EVENTS {
217 P4_EVENT_TC_DELIVER_MODE,
218 P4_EVENT_BPU_FETCH_REQUEST,
219 P4_EVENT_ITLB_REFERENCE,
220 P4_EVENT_MEMORY_CANCEL,
221 P4_EVENT_MEMORY_COMPLETE,
222 P4_EVENT_LOAD_PORT_REPLAY,
223 P4_EVENT_STORE_PORT_REPLAY,
224 P4_EVENT_MOB_LOAD_REPLAY,
225 P4_EVENT_PAGE_WALK_TYPE,
226 P4_EVENT_BSQ_CACHE_REFERENCE,
227 P4_EVENT_IOQ_ALLOCATION,
228 P4_EVENT_IOQ_ACTIVE_ENTRIES,
229 P4_EVENT_FSB_DATA_ACTIVITY,
230 P4_EVENT_BSQ_ALLOCATION,
231 P4_EVENT_BSQ_ACTIVE_ENTRIES,
232 P4_EVENT_SSE_INPUT_ASSIST,
233 P4_EVENT_PACKED_SP_UOP,
234 P4_EVENT_PACKED_DP_UOP,
235 P4_EVENT_SCALAR_SP_UOP,
236 P4_EVENT_SCALAR_DP_UOP,
237 P4_EVENT_64BIT_MMX_UOP,
238 P4_EVENT_128BIT_MMX_UOP,
239 P4_EVENT_X87_FP_UOP,
240 P4_EVENT_TC_MISC,
241 P4_EVENT_GLOBAL_POWER_EVENTS,
242 P4_EVENT_TC_MS_XFER,
243 P4_EVENT_UOP_QUEUE_WRITES,
244 P4_EVENT_RETIRED_MISPRED_BRANCH_TYPE,
245 P4_EVENT_RETIRED_BRANCH_TYPE,
246 P4_EVENT_RESOURCE_STALL,
247 P4_EVENT_WC_BUFFER,
248 P4_EVENT_B2B_CYCLES,
249 P4_EVENT_BNR,
250 P4_EVENT_SNOOP,
251 P4_EVENT_RESPONSE,
252 P4_EVENT_FRONT_END_EVENT,
253 P4_EVENT_EXECUTION_EVENT,
254 P4_EVENT_REPLAY_EVENT,
255 P4_EVENT_INSTR_RETIRED,
256 P4_EVENT_UOPS_RETIRED,
257 P4_EVENT_UOP_TYPE,
258 P4_EVENT_BRANCH_RETIRED,
259 P4_EVENT_MISPRED_BRANCH_RETIRED,
260 P4_EVENT_X87_ASSIST,
261 P4_EVENT_MACHINE_CLEAR,
262 P4_EVENT_INSTR_COMPLETED,
263};
264
265#define P4_OPCODE(event) event##_OPCODE
266#define P4_OPCODE_ESEL(opcode) ((opcode & 0x00ff) >> 0)
267#define P4_OPCODE_EVNT(opcode) ((opcode & 0xff00) >> 8)
268#define P4_OPCODE_PACK(event, sel) (((event) << 8) | sel)
269
270/*
271 * Comments below the event represent ESCR restriction
272 * for this event and counter index per ESCR
273 *
274 * MSR_P4_IQ_ESCR0 and MSR_P4_IQ_ESCR1 are available only on early
275 * processor builds (family 0FH, models 01H-02H). These MSRs
276 * are not available on later versions, so that we don't use
277 * them completely
278 *
279 * Also note that CCCR1 do not have P4_CCCR_ENABLE bit properly
280 * working so that we should not use this CCCR and respective
281 * counter as result
282 */
283enum P4_EVENT_OPCODES {
284 P4_OPCODE(P4_EVENT_TC_DELIVER_MODE) = P4_OPCODE_PACK(0x01, 0x01),
285 /*
286 * MSR_P4_TC_ESCR0: 4, 5
287 * MSR_P4_TC_ESCR1: 6, 7
288 */
289
290 P4_OPCODE(P4_EVENT_BPU_FETCH_REQUEST) = P4_OPCODE_PACK(0x03, 0x00),
291 /*
292 * MSR_P4_BPU_ESCR0: 0, 1
293 * MSR_P4_BPU_ESCR1: 2, 3
294 */
295
296 P4_OPCODE(P4_EVENT_ITLB_REFERENCE) = P4_OPCODE_PACK(0x18, 0x03),
297 /*
298 * MSR_P4_ITLB_ESCR0: 0, 1
299 * MSR_P4_ITLB_ESCR1: 2, 3
300 */
301
302 P4_OPCODE(P4_EVENT_MEMORY_CANCEL) = P4_OPCODE_PACK(0x02, 0x05),
303 /*
304 * MSR_P4_DAC_ESCR0: 8, 9
305 * MSR_P4_DAC_ESCR1: 10, 11
306 */
307
308 P4_OPCODE(P4_EVENT_MEMORY_COMPLETE) = P4_OPCODE_PACK(0x08, 0x02),
309 /*
310 * MSR_P4_SAAT_ESCR0: 8, 9
311 * MSR_P4_SAAT_ESCR1: 10, 11
312 */
313
314 P4_OPCODE(P4_EVENT_LOAD_PORT_REPLAY) = P4_OPCODE_PACK(0x04, 0x02),
315 /*
316 * MSR_P4_SAAT_ESCR0: 8, 9
317 * MSR_P4_SAAT_ESCR1: 10, 11
318 */
319
320 P4_OPCODE(P4_EVENT_STORE_PORT_REPLAY) = P4_OPCODE_PACK(0x05, 0x02),
321 /*
322 * MSR_P4_SAAT_ESCR0: 8, 9
323 * MSR_P4_SAAT_ESCR1: 10, 11
324 */
325
326 P4_OPCODE(P4_EVENT_MOB_LOAD_REPLAY) = P4_OPCODE_PACK(0x03, 0x02),
327 /*
328 * MSR_P4_MOB_ESCR0: 0, 1
329 * MSR_P4_MOB_ESCR1: 2, 3
330 */
331
332 P4_OPCODE(P4_EVENT_PAGE_WALK_TYPE) = P4_OPCODE_PACK(0x01, 0x04),
333 /*
334 * MSR_P4_PMH_ESCR0: 0, 1
335 * MSR_P4_PMH_ESCR1: 2, 3
336 */
337
338 P4_OPCODE(P4_EVENT_BSQ_CACHE_REFERENCE) = P4_OPCODE_PACK(0x0c, 0x07),
339 /*
340 * MSR_P4_BSU_ESCR0: 0, 1
341 * MSR_P4_BSU_ESCR1: 2, 3
342 */
343
344 P4_OPCODE(P4_EVENT_IOQ_ALLOCATION) = P4_OPCODE_PACK(0x03, 0x06),
345 /*
346 * MSR_P4_FSB_ESCR0: 0, 1
347 * MSR_P4_FSB_ESCR1: 2, 3
348 */
349
350 P4_OPCODE(P4_EVENT_IOQ_ACTIVE_ENTRIES) = P4_OPCODE_PACK(0x1a, 0x06),
351 /*
352 * MSR_P4_FSB_ESCR1: 2, 3
353 */
354
355 P4_OPCODE(P4_EVENT_FSB_DATA_ACTIVITY) = P4_OPCODE_PACK(0x17, 0x06),
356 /*
357 * MSR_P4_FSB_ESCR0: 0, 1
358 * MSR_P4_FSB_ESCR1: 2, 3
359 */
360
361 P4_OPCODE(P4_EVENT_BSQ_ALLOCATION) = P4_OPCODE_PACK(0x05, 0x07),
362 /*
363 * MSR_P4_BSU_ESCR0: 0, 1
364 */
365
366 P4_OPCODE(P4_EVENT_BSQ_ACTIVE_ENTRIES) = P4_OPCODE_PACK(0x06, 0x07),
367 /*
368 * NOTE: no ESCR name in docs, it's guessed
369 * MSR_P4_BSU_ESCR1: 2, 3
370 */
371
372 P4_OPCODE(P4_EVENT_SSE_INPUT_ASSIST) = P4_OPCODE_PACK(0x34, 0x01),
373 /*
374 * MSR_P4_FIRM_ESCR0: 8, 9
375 * MSR_P4_FIRM_ESCR1: 10, 11
376 */
377
378 P4_OPCODE(P4_EVENT_PACKED_SP_UOP) = P4_OPCODE_PACK(0x08, 0x01),
379 /*
380 * MSR_P4_FIRM_ESCR0: 8, 9
381 * MSR_P4_FIRM_ESCR1: 10, 11
382 */
383
384 P4_OPCODE(P4_EVENT_PACKED_DP_UOP) = P4_OPCODE_PACK(0x0c, 0x01),
385 /*
386 * MSR_P4_FIRM_ESCR0: 8, 9
387 * MSR_P4_FIRM_ESCR1: 10, 11
388 */
389
390 P4_OPCODE(P4_EVENT_SCALAR_SP_UOP) = P4_OPCODE_PACK(0x0a, 0x01),
391 /*
392 * MSR_P4_FIRM_ESCR0: 8, 9
393 * MSR_P4_FIRM_ESCR1: 10, 11
394 */
395
396 P4_OPCODE(P4_EVENT_SCALAR_DP_UOP) = P4_OPCODE_PACK(0x0e, 0x01),
397 /*
398 * MSR_P4_FIRM_ESCR0: 8, 9
399 * MSR_P4_FIRM_ESCR1: 10, 11
400 */
401
402 P4_OPCODE(P4_EVENT_64BIT_MMX_UOP) = P4_OPCODE_PACK(0x02, 0x01),
403 /*
404 * MSR_P4_FIRM_ESCR0: 8, 9
405 * MSR_P4_FIRM_ESCR1: 10, 11
406 */
407
408 P4_OPCODE(P4_EVENT_128BIT_MMX_UOP) = P4_OPCODE_PACK(0x1a, 0x01),
409 /*
410 * MSR_P4_FIRM_ESCR0: 8, 9
411 * MSR_P4_FIRM_ESCR1: 10, 11
412 */
413
414 P4_OPCODE(P4_EVENT_X87_FP_UOP) = P4_OPCODE_PACK(0x04, 0x01),
415 /*
416 * MSR_P4_FIRM_ESCR0: 8, 9
417 * MSR_P4_FIRM_ESCR1: 10, 11
418 */
419
420 P4_OPCODE(P4_EVENT_TC_MISC) = P4_OPCODE_PACK(0x06, 0x01),
421 /*
422 * MSR_P4_TC_ESCR0: 4, 5
423 * MSR_P4_TC_ESCR1: 6, 7
424 */
425
426 P4_OPCODE(P4_EVENT_GLOBAL_POWER_EVENTS) = P4_OPCODE_PACK(0x13, 0x06),
427 /*
428 * MSR_P4_FSB_ESCR0: 0, 1
429 * MSR_P4_FSB_ESCR1: 2, 3
430 */
431
432 P4_OPCODE(P4_EVENT_TC_MS_XFER) = P4_OPCODE_PACK(0x05, 0x00),
433 /*
434 * MSR_P4_MS_ESCR0: 4, 5
435 * MSR_P4_MS_ESCR1: 6, 7
436 */
437
438 P4_OPCODE(P4_EVENT_UOP_QUEUE_WRITES) = P4_OPCODE_PACK(0x09, 0x00),
439 /*
440 * MSR_P4_MS_ESCR0: 4, 5
441 * MSR_P4_MS_ESCR1: 6, 7
442 */
443
444 P4_OPCODE(P4_EVENT_RETIRED_MISPRED_BRANCH_TYPE) = P4_OPCODE_PACK(0x05, 0x02),
445 /*
446 * MSR_P4_TBPU_ESCR0: 4, 5
447 * MSR_P4_TBPU_ESCR1: 6, 7
448 */
449
450 P4_OPCODE(P4_EVENT_RETIRED_BRANCH_TYPE) = P4_OPCODE_PACK(0x04, 0x02),
451 /*
452 * MSR_P4_TBPU_ESCR0: 4, 5
453 * MSR_P4_TBPU_ESCR1: 6, 7
454 */
455
456 P4_OPCODE(P4_EVENT_RESOURCE_STALL) = P4_OPCODE_PACK(0x01, 0x01),
457 /*
458 * MSR_P4_ALF_ESCR0: 12, 13, 16
459 * MSR_P4_ALF_ESCR1: 14, 15, 17
460 */
461
462 P4_OPCODE(P4_EVENT_WC_BUFFER) = P4_OPCODE_PACK(0x05, 0x05),
463 /*
464 * MSR_P4_DAC_ESCR0: 8, 9
465 * MSR_P4_DAC_ESCR1: 10, 11
466 */
467
468 P4_OPCODE(P4_EVENT_B2B_CYCLES) = P4_OPCODE_PACK(0x16, 0x03),
469 /*
470 * MSR_P4_FSB_ESCR0: 0, 1
471 * MSR_P4_FSB_ESCR1: 2, 3
472 */
473
474 P4_OPCODE(P4_EVENT_BNR) = P4_OPCODE_PACK(0x08, 0x03),
475 /*
476 * MSR_P4_FSB_ESCR0: 0, 1
477 * MSR_P4_FSB_ESCR1: 2, 3
478 */
479
480 P4_OPCODE(P4_EVENT_SNOOP) = P4_OPCODE_PACK(0x06, 0x03),
481 /*
482 * MSR_P4_FSB_ESCR0: 0, 1
483 * MSR_P4_FSB_ESCR1: 2, 3
484 */
485
486 P4_OPCODE(P4_EVENT_RESPONSE) = P4_OPCODE_PACK(0x04, 0x03),
487 /*
488 * MSR_P4_FSB_ESCR0: 0, 1
489 * MSR_P4_FSB_ESCR1: 2, 3
490 */
491
492 P4_OPCODE(P4_EVENT_FRONT_END_EVENT) = P4_OPCODE_PACK(0x08, 0x05),
493 /*
494 * MSR_P4_CRU_ESCR2: 12, 13, 16
495 * MSR_P4_CRU_ESCR3: 14, 15, 17
496 */
497
498 P4_OPCODE(P4_EVENT_EXECUTION_EVENT) = P4_OPCODE_PACK(0x0c, 0x05),
499 /*
500 * MSR_P4_CRU_ESCR2: 12, 13, 16
501 * MSR_P4_CRU_ESCR3: 14, 15, 17
502 */
503
504 P4_OPCODE(P4_EVENT_REPLAY_EVENT) = P4_OPCODE_PACK(0x09, 0x05),
505 /*
506 * MSR_P4_CRU_ESCR2: 12, 13, 16
507 * MSR_P4_CRU_ESCR3: 14, 15, 17
508 */
509
510 P4_OPCODE(P4_EVENT_INSTR_RETIRED) = P4_OPCODE_PACK(0x02, 0x04),
511 /*
512 * MSR_P4_CRU_ESCR0: 12, 13, 16
513 * MSR_P4_CRU_ESCR1: 14, 15, 17
514 */
515
516 P4_OPCODE(P4_EVENT_UOPS_RETIRED) = P4_OPCODE_PACK(0x01, 0x04),
517 /*
518 * MSR_P4_CRU_ESCR0: 12, 13, 16
519 * MSR_P4_CRU_ESCR1: 14, 15, 17
520 */
521
522 P4_OPCODE(P4_EVENT_UOP_TYPE) = P4_OPCODE_PACK(0x02, 0x02),
523 /*
524 * MSR_P4_RAT_ESCR0: 12, 13, 16
525 * MSR_P4_RAT_ESCR1: 14, 15, 17
526 */
527
528 P4_OPCODE(P4_EVENT_BRANCH_RETIRED) = P4_OPCODE_PACK(0x06, 0x05),
529 /*
530 * MSR_P4_CRU_ESCR2: 12, 13, 16
531 * MSR_P4_CRU_ESCR3: 14, 15, 17
532 */
533
534 P4_OPCODE(P4_EVENT_MISPRED_BRANCH_RETIRED) = P4_OPCODE_PACK(0x03, 0x04),
535 /*
536 * MSR_P4_CRU_ESCR0: 12, 13, 16
537 * MSR_P4_CRU_ESCR1: 14, 15, 17
538 */
539
540 P4_OPCODE(P4_EVENT_X87_ASSIST) = P4_OPCODE_PACK(0x03, 0x05),
541 /*
542 * MSR_P4_CRU_ESCR2: 12, 13, 16
543 * MSR_P4_CRU_ESCR3: 14, 15, 17
544 */
545
546 P4_OPCODE(P4_EVENT_MACHINE_CLEAR) = P4_OPCODE_PACK(0x02, 0x05),
547 /*
548 * MSR_P4_CRU_ESCR2: 12, 13, 16
549 * MSR_P4_CRU_ESCR3: 14, 15, 17
550 */
551
552 P4_OPCODE(P4_EVENT_INSTR_COMPLETED) = P4_OPCODE_PACK(0x07, 0x04),
553 /*
554 * MSR_P4_CRU_ESCR0: 12, 13, 16
555 * MSR_P4_CRU_ESCR1: 14, 15, 17
556 */
557};
558
559/*
560 * a caller should use P4_ESCR_EMASK_NAME helper to
561 * pick the EventMask needed, for example
562 *
563 * P4_ESCR_EMASK_NAME(P4_EVENT_TC_DELIVER_MODE, DD)
564 */
565enum P4_ESCR_EMASKS {
566 P4_GEN_ESCR_EMASK(P4_EVENT_TC_DELIVER_MODE, DD, 0),
567 P4_GEN_ESCR_EMASK(P4_EVENT_TC_DELIVER_MODE, DB, 1),
568 P4_GEN_ESCR_EMASK(P4_EVENT_TC_DELIVER_MODE, DI, 2),
569 P4_GEN_ESCR_EMASK(P4_EVENT_TC_DELIVER_MODE, BD, 3),
570 P4_GEN_ESCR_EMASK(P4_EVENT_TC_DELIVER_MODE, BB, 4),
571 P4_GEN_ESCR_EMASK(P4_EVENT_TC_DELIVER_MODE, BI, 5),
572 P4_GEN_ESCR_EMASK(P4_EVENT_TC_DELIVER_MODE, ID, 6),
573
574 P4_GEN_ESCR_EMASK(P4_EVENT_BPU_FETCH_REQUEST, TCMISS, 0),
575
576 P4_GEN_ESCR_EMASK(P4_EVENT_ITLB_REFERENCE, HIT, 0),
577 P4_GEN_ESCR_EMASK(P4_EVENT_ITLB_REFERENCE, MISS, 1),
578 P4_GEN_ESCR_EMASK(P4_EVENT_ITLB_REFERENCE, HIT_UK, 2),
579
580 P4_GEN_ESCR_EMASK(P4_EVENT_MEMORY_CANCEL, ST_RB_FULL, 2),
581 P4_GEN_ESCR_EMASK(P4_EVENT_MEMORY_CANCEL, 64K_CONF, 3),
582
583 P4_GEN_ESCR_EMASK(P4_EVENT_MEMORY_COMPLETE, LSC, 0),
584 P4_GEN_ESCR_EMASK(P4_EVENT_MEMORY_COMPLETE, SSC, 1),
585
586 P4_GEN_ESCR_EMASK(P4_EVENT_LOAD_PORT_REPLAY, SPLIT_LD, 1),
587
588 P4_GEN_ESCR_EMASK(P4_EVENT_STORE_PORT_REPLAY, SPLIT_ST, 1),
589
590 P4_GEN_ESCR_EMASK(P4_EVENT_MOB_LOAD_REPLAY, NO_STA, 1),
591 P4_GEN_ESCR_EMASK(P4_EVENT_MOB_LOAD_REPLAY, NO_STD, 3),
592 P4_GEN_ESCR_EMASK(P4_EVENT_MOB_LOAD_REPLAY, PARTIAL_DATA, 4),
593 P4_GEN_ESCR_EMASK(P4_EVENT_MOB_LOAD_REPLAY, UNALGN_ADDR, 5),
594
595 P4_GEN_ESCR_EMASK(P4_EVENT_PAGE_WALK_TYPE, DTMISS, 0),
596 P4_GEN_ESCR_EMASK(P4_EVENT_PAGE_WALK_TYPE, ITMISS, 1),
597
598 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_CACHE_REFERENCE, RD_2ndL_HITS, 0),
599 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_CACHE_REFERENCE, RD_2ndL_HITE, 1),
600 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_CACHE_REFERENCE, RD_2ndL_HITM, 2),
601 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_CACHE_REFERENCE, RD_3rdL_HITS, 3),
602 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_CACHE_REFERENCE, RD_3rdL_HITE, 4),
603 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_CACHE_REFERENCE, RD_3rdL_HITM, 5),
604 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_CACHE_REFERENCE, RD_2ndL_MISS, 8),
605 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_CACHE_REFERENCE, RD_3rdL_MISS, 9),
606 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_CACHE_REFERENCE, WR_2ndL_MISS, 10),
607
608 P4_GEN_ESCR_EMASK(P4_EVENT_IOQ_ALLOCATION, DEFAULT, 0),
609 P4_GEN_ESCR_EMASK(P4_EVENT_IOQ_ALLOCATION, ALL_READ, 5),
610 P4_GEN_ESCR_EMASK(P4_EVENT_IOQ_ALLOCATION, ALL_WRITE, 6),
611 P4_GEN_ESCR_EMASK(P4_EVENT_IOQ_ALLOCATION, MEM_UC, 7),
612 P4_GEN_ESCR_EMASK(P4_EVENT_IOQ_ALLOCATION, MEM_WC, 8),
613 P4_GEN_ESCR_EMASK(P4_EVENT_IOQ_ALLOCATION, MEM_WT, 9),
614 P4_GEN_ESCR_EMASK(P4_EVENT_IOQ_ALLOCATION, MEM_WP, 10),
615 P4_GEN_ESCR_EMASK(P4_EVENT_IOQ_ALLOCATION, MEM_WB, 11),
616 P4_GEN_ESCR_EMASK(P4_EVENT_IOQ_ALLOCATION, OWN, 13),
617 P4_GEN_ESCR_EMASK(P4_EVENT_IOQ_ALLOCATION, OTHER, 14),
618 P4_GEN_ESCR_EMASK(P4_EVENT_IOQ_ALLOCATION, PREFETCH, 15),
619
620 P4_GEN_ESCR_EMASK(P4_EVENT_IOQ_ACTIVE_ENTRIES, DEFAULT, 0),
621 P4_GEN_ESCR_EMASK(P4_EVENT_IOQ_ACTIVE_ENTRIES, ALL_READ, 5),
622 P4_GEN_ESCR_EMASK(P4_EVENT_IOQ_ACTIVE_ENTRIES, ALL_WRITE, 6),
623 P4_GEN_ESCR_EMASK(P4_EVENT_IOQ_ACTIVE_ENTRIES, MEM_UC, 7),
624 P4_GEN_ESCR_EMASK(P4_EVENT_IOQ_ACTIVE_ENTRIES, MEM_WC, 8),
625 P4_GEN_ESCR_EMASK(P4_EVENT_IOQ_ACTIVE_ENTRIES, MEM_WT, 9),
626 P4_GEN_ESCR_EMASK(P4_EVENT_IOQ_ACTIVE_ENTRIES, MEM_WP, 10),
627 P4_GEN_ESCR_EMASK(P4_EVENT_IOQ_ACTIVE_ENTRIES, MEM_WB, 11),
628 P4_GEN_ESCR_EMASK(P4_EVENT_IOQ_ACTIVE_ENTRIES, OWN, 13),
629 P4_GEN_ESCR_EMASK(P4_EVENT_IOQ_ACTIVE_ENTRIES, OTHER, 14),
630 P4_GEN_ESCR_EMASK(P4_EVENT_IOQ_ACTIVE_ENTRIES, PREFETCH, 15),
631
632 P4_GEN_ESCR_EMASK(P4_EVENT_FSB_DATA_ACTIVITY, DRDY_DRV, 0),
633 P4_GEN_ESCR_EMASK(P4_EVENT_FSB_DATA_ACTIVITY, DRDY_OWN, 1),
634 P4_GEN_ESCR_EMASK(P4_EVENT_FSB_DATA_ACTIVITY, DRDY_OTHER, 2),
635 P4_GEN_ESCR_EMASK(P4_EVENT_FSB_DATA_ACTIVITY, DBSY_DRV, 3),
636 P4_GEN_ESCR_EMASK(P4_EVENT_FSB_DATA_ACTIVITY, DBSY_OWN, 4),
637 P4_GEN_ESCR_EMASK(P4_EVENT_FSB_DATA_ACTIVITY, DBSY_OTHER, 5),
638
639 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ALLOCATION, REQ_TYPE0, 0),
640 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ALLOCATION, REQ_TYPE1, 1),
641 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ALLOCATION, REQ_LEN0, 2),
642 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ALLOCATION, REQ_LEN1, 3),
643 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ALLOCATION, REQ_IO_TYPE, 5),
644 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ALLOCATION, REQ_LOCK_TYPE, 6),
645 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ALLOCATION, REQ_CACHE_TYPE, 7),
646 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ALLOCATION, REQ_SPLIT_TYPE, 8),
647 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ALLOCATION, REQ_DEM_TYPE, 9),
648 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ALLOCATION, REQ_ORD_TYPE, 10),
649 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ALLOCATION, MEM_TYPE0, 11),
650 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ALLOCATION, MEM_TYPE1, 12),
651 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ALLOCATION, MEM_TYPE2, 13),
652
653 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ACTIVE_ENTRIES, REQ_TYPE0, 0),
654 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ACTIVE_ENTRIES, REQ_TYPE1, 1),
655 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ACTIVE_ENTRIES, REQ_LEN0, 2),
656 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ACTIVE_ENTRIES, REQ_LEN1, 3),
657 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ACTIVE_ENTRIES, REQ_IO_TYPE, 5),
658 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ACTIVE_ENTRIES, REQ_LOCK_TYPE, 6),
659 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ACTIVE_ENTRIES, REQ_CACHE_TYPE, 7),
660 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ACTIVE_ENTRIES, REQ_SPLIT_TYPE, 8),
661 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ACTIVE_ENTRIES, REQ_DEM_TYPE, 9),
662 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ACTIVE_ENTRIES, REQ_ORD_TYPE, 10),
663 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ACTIVE_ENTRIES, MEM_TYPE0, 11),
664 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ACTIVE_ENTRIES, MEM_TYPE1, 12),
665 P4_GEN_ESCR_EMASK(P4_EVENT_BSQ_ACTIVE_ENTRIES, MEM_TYPE2, 13),
666
667 P4_GEN_ESCR_EMASK(P4_EVENT_SSE_INPUT_ASSIST, ALL, 15),
668
669 P4_GEN_ESCR_EMASK(P4_EVENT_PACKED_SP_UOP, ALL, 15),
670
671 P4_GEN_ESCR_EMASK(P4_EVENT_PACKED_DP_UOP, ALL, 15),
672
673 P4_GEN_ESCR_EMASK(P4_EVENT_SCALAR_SP_UOP, ALL, 15),
674
675 P4_GEN_ESCR_EMASK(P4_EVENT_SCALAR_DP_UOP, ALL, 15),
676
677 P4_GEN_ESCR_EMASK(P4_EVENT_64BIT_MMX_UOP, ALL, 15),
678
679 P4_GEN_ESCR_EMASK(P4_EVENT_128BIT_MMX_UOP, ALL, 15),
680
681 P4_GEN_ESCR_EMASK(P4_EVENT_X87_FP_UOP, ALL, 15),
682
683 P4_GEN_ESCR_EMASK(P4_EVENT_TC_MISC, FLUSH, 4),
684
685 P4_GEN_ESCR_EMASK(P4_EVENT_GLOBAL_POWER_EVENTS, RUNNING, 0),
686
687 P4_GEN_ESCR_EMASK(P4_EVENT_TC_MS_XFER, CISC, 0),
688
689 P4_GEN_ESCR_EMASK(P4_EVENT_UOP_QUEUE_WRITES, FROM_TC_BUILD, 0),
690 P4_GEN_ESCR_EMASK(P4_EVENT_UOP_QUEUE_WRITES, FROM_TC_DELIVER, 1),
691 P4_GEN_ESCR_EMASK(P4_EVENT_UOP_QUEUE_WRITES, FROM_ROM, 2),
692
693 P4_GEN_ESCR_EMASK(P4_EVENT_RETIRED_MISPRED_BRANCH_TYPE, CONDITIONAL, 1),
694 P4_GEN_ESCR_EMASK(P4_EVENT_RETIRED_MISPRED_BRANCH_TYPE, CALL, 2),
695 P4_GEN_ESCR_EMASK(P4_EVENT_RETIRED_MISPRED_BRANCH_TYPE, RETURN, 3),
696 P4_GEN_ESCR_EMASK(P4_EVENT_RETIRED_MISPRED_BRANCH_TYPE, INDIRECT, 4),
697
698 P4_GEN_ESCR_EMASK(P4_EVENT_RETIRED_BRANCH_TYPE, CONDITIONAL, 1),
699 P4_GEN_ESCR_EMASK(P4_EVENT_RETIRED_BRANCH_TYPE, CALL, 2),
700 P4_GEN_ESCR_EMASK(P4_EVENT_RETIRED_BRANCH_TYPE, RETURN, 3),
701 P4_GEN_ESCR_EMASK(P4_EVENT_RETIRED_BRANCH_TYPE, INDIRECT, 4),
702
703 P4_GEN_ESCR_EMASK(P4_EVENT_RESOURCE_STALL, SBFULL, 5),
704
705 P4_GEN_ESCR_EMASK(P4_EVENT_WC_BUFFER, WCB_EVICTS, 0),
706 P4_GEN_ESCR_EMASK(P4_EVENT_WC_BUFFER, WCB_FULL_EVICTS, 1),
707
708 P4_GEN_ESCR_EMASK(P4_EVENT_FRONT_END_EVENT, NBOGUS, 0),
709 P4_GEN_ESCR_EMASK(P4_EVENT_FRONT_END_EVENT, BOGUS, 1),
710
711 P4_GEN_ESCR_EMASK(P4_EVENT_EXECUTION_EVENT, NBOGUS0, 0),
712 P4_GEN_ESCR_EMASK(P4_EVENT_EXECUTION_EVENT, NBOGUS1, 1),
713 P4_GEN_ESCR_EMASK(P4_EVENT_EXECUTION_EVENT, NBOGUS2, 2),
714 P4_GEN_ESCR_EMASK(P4_EVENT_EXECUTION_EVENT, NBOGUS3, 3),
715 P4_GEN_ESCR_EMASK(P4_EVENT_EXECUTION_EVENT, BOGUS0, 4),
716 P4_GEN_ESCR_EMASK(P4_EVENT_EXECUTION_EVENT, BOGUS1, 5),
717 P4_GEN_ESCR_EMASK(P4_EVENT_EXECUTION_EVENT, BOGUS2, 6),
718 P4_GEN_ESCR_EMASK(P4_EVENT_EXECUTION_EVENT, BOGUS3, 7),
719
720 P4_GEN_ESCR_EMASK(P4_EVENT_REPLAY_EVENT, NBOGUS, 0),
721 P4_GEN_ESCR_EMASK(P4_EVENT_REPLAY_EVENT, BOGUS, 1),
722
723 P4_GEN_ESCR_EMASK(P4_EVENT_INSTR_RETIRED, NBOGUSNTAG, 0),
724 P4_GEN_ESCR_EMASK(P4_EVENT_INSTR_RETIRED, NBOGUSTAG, 1),
725 P4_GEN_ESCR_EMASK(P4_EVENT_INSTR_RETIRED, BOGUSNTAG, 2),
726 P4_GEN_ESCR_EMASK(P4_EVENT_INSTR_RETIRED, BOGUSTAG, 3),
727
728 P4_GEN_ESCR_EMASK(P4_EVENT_UOPS_RETIRED, NBOGUS, 0),
729 P4_GEN_ESCR_EMASK(P4_EVENT_UOPS_RETIRED, BOGUS, 1),
730
731 P4_GEN_ESCR_EMASK(P4_EVENT_UOP_TYPE, TAGLOADS, 1),
732 P4_GEN_ESCR_EMASK(P4_EVENT_UOP_TYPE, TAGSTORES, 2),
733
734 P4_GEN_ESCR_EMASK(P4_EVENT_BRANCH_RETIRED, MMNP, 0),
735 P4_GEN_ESCR_EMASK(P4_EVENT_BRANCH_RETIRED, MMNM, 1),
736 P4_GEN_ESCR_EMASK(P4_EVENT_BRANCH_RETIRED, MMTP, 2),
737 P4_GEN_ESCR_EMASK(P4_EVENT_BRANCH_RETIRED, MMTM, 3),
738
739 P4_GEN_ESCR_EMASK(P4_EVENT_MISPRED_BRANCH_RETIRED, NBOGUS, 0),
740
741 P4_GEN_ESCR_EMASK(P4_EVENT_X87_ASSIST, FPSU, 0),
742 P4_GEN_ESCR_EMASK(P4_EVENT_X87_ASSIST, FPSO, 1),
743 P4_GEN_ESCR_EMASK(P4_EVENT_X87_ASSIST, POAO, 2),
744 P4_GEN_ESCR_EMASK(P4_EVENT_X87_ASSIST, POAU, 3),
745 P4_GEN_ESCR_EMASK(P4_EVENT_X87_ASSIST, PREA, 4),
746
747 P4_GEN_ESCR_EMASK(P4_EVENT_MACHINE_CLEAR, CLEAR, 0),
748 P4_GEN_ESCR_EMASK(P4_EVENT_MACHINE_CLEAR, MOCLEAR, 1),
749 P4_GEN_ESCR_EMASK(P4_EVENT_MACHINE_CLEAR, SMCLEAR, 2),
750
751 P4_GEN_ESCR_EMASK(P4_EVENT_INSTR_COMPLETED, NBOGUS, 0),
752 P4_GEN_ESCR_EMASK(P4_EVENT_INSTR_COMPLETED, BOGUS, 1),
753};
754
755/* P4 PEBS: stale for a while */
756#define P4_PEBS_METRIC_MASK 0x00001fffU
757#define P4_PEBS_UOB_TAG 0x01000000U
758#define P4_PEBS_ENABLE 0x02000000U
759
760/* Replay metrics for MSR_IA32_PEBS_ENABLE and MSR_P4_PEBS_MATRIX_VERT */
761#define P4_PEBS__1stl_cache_load_miss_retired 0x3000001
762#define P4_PEBS__2ndl_cache_load_miss_retired 0x3000002
763#define P4_PEBS__dtlb_load_miss_retired 0x3000004
764#define P4_PEBS__dtlb_store_miss_retired 0x3000004
765#define P4_PEBS__dtlb_all_miss_retired 0x3000004
766#define P4_PEBS__tagged_mispred_branch 0x3018000
767#define P4_PEBS__mob_load_replay_retired 0x3000200
768#define P4_PEBS__split_load_retired 0x3000400
769#define P4_PEBS__split_store_retired 0x3000400
770
771#define P4_VERT__1stl_cache_load_miss_retired 0x0000001
772#define P4_VERT__2ndl_cache_load_miss_retired 0x0000001
773#define P4_VERT__dtlb_load_miss_retired 0x0000001
774#define P4_VERT__dtlb_store_miss_retired 0x0000002
775#define P4_VERT__dtlb_all_miss_retired 0x0000003
776#define P4_VERT__tagged_mispred_branch 0x0000010
777#define P4_VERT__mob_load_replay_retired 0x0000001
778#define P4_VERT__split_load_retired 0x0000001
779#define P4_VERT__split_store_retired 0x0000002
780
781enum P4_CACHE_EVENTS {
782 P4_CACHE__NONE,
783
784 P4_CACHE__1stl_cache_load_miss_retired,
785 P4_CACHE__2ndl_cache_load_miss_retired,
786 P4_CACHE__dtlb_load_miss_retired,
787 P4_CACHE__dtlb_store_miss_retired,
788 P4_CACHE__itlb_reference_hit,
789 P4_CACHE__itlb_reference_miss,
790
791 P4_CACHE__MAX
792};
793
794#endif /* PERF_EVENT_P4_H */
diff --git a/arch/x86/include/asm/processor.h b/arch/x86/include/asm/processor.h
index b753ea59703a..32428b410b55 100644
--- a/arch/x86/include/asm/processor.h
+++ b/arch/x86/include/asm/processor.h
@@ -21,7 +21,6 @@ struct mm_struct;
21#include <asm/msr.h> 21#include <asm/msr.h>
22#include <asm/desc_defs.h> 22#include <asm/desc_defs.h>
23#include <asm/nops.h> 23#include <asm/nops.h>
24#include <asm/ds.h>
25 24
26#include <linux/personality.h> 25#include <linux/personality.h>
27#include <linux/cpumask.h> 26#include <linux/cpumask.h>
@@ -29,6 +28,7 @@ struct mm_struct;
29#include <linux/threads.h> 28#include <linux/threads.h>
30#include <linux/math64.h> 29#include <linux/math64.h>
31#include <linux/init.h> 30#include <linux/init.h>
31#include <linux/err.h>
32 32
33#define HBP_NUM 4 33#define HBP_NUM 4
34/* 34/*
@@ -473,10 +473,6 @@ struct thread_struct {
473 unsigned long iopl; 473 unsigned long iopl;
474 /* Max allowed port in the bitmap, in bytes: */ 474 /* Max allowed port in the bitmap, in bytes: */
475 unsigned io_bitmap_max; 475 unsigned io_bitmap_max;
476/* MSR_IA32_DEBUGCTLMSR value to switch in if TIF_DEBUGCTLMSR is set. */
477 unsigned long debugctlmsr;
478 /* Debug Store context; see asm/ds.h */
479 struct ds_context *ds_ctx;
480}; 476};
481 477
482static inline unsigned long native_get_debugreg(int regno) 478static inline unsigned long native_get_debugreg(int regno)
@@ -803,7 +799,7 @@ extern void cpu_init(void);
803 799
804static inline unsigned long get_debugctlmsr(void) 800static inline unsigned long get_debugctlmsr(void)
805{ 801{
806 unsigned long debugctlmsr = 0; 802 unsigned long debugctlmsr = 0;
807 803
808#ifndef CONFIG_X86_DEBUGCTLMSR 804#ifndef CONFIG_X86_DEBUGCTLMSR
809 if (boot_cpu_data.x86 < 6) 805 if (boot_cpu_data.x86 < 6)
@@ -811,21 +807,6 @@ static inline unsigned long get_debugctlmsr(void)
811#endif 807#endif
812 rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr); 808 rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
813 809
814 return debugctlmsr;
815}
816
817static inline unsigned long get_debugctlmsr_on_cpu(int cpu)
818{
819 u64 debugctlmsr = 0;
820 u32 val1, val2;
821
822#ifndef CONFIG_X86_DEBUGCTLMSR
823 if (boot_cpu_data.x86 < 6)
824 return 0;
825#endif
826 rdmsr_on_cpu(cpu, MSR_IA32_DEBUGCTLMSR, &val1, &val2);
827 debugctlmsr = val1 | ((u64)val2 << 32);
828
829 return debugctlmsr; 810 return debugctlmsr;
830} 811}
831 812
@@ -838,18 +819,6 @@ static inline void update_debugctlmsr(unsigned long debugctlmsr)
838 wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr); 819 wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
839} 820}
840 821
841static inline void update_debugctlmsr_on_cpu(int cpu,
842 unsigned long debugctlmsr)
843{
844#ifndef CONFIG_X86_DEBUGCTLMSR
845 if (boot_cpu_data.x86 < 6)
846 return;
847#endif
848 wrmsr_on_cpu(cpu, MSR_IA32_DEBUGCTLMSR,
849 (u32)((u64)debugctlmsr),
850 (u32)((u64)debugctlmsr >> 32));
851}
852
853/* 822/*
854 * from system description table in BIOS. Mostly for MCA use, but 823 * from system description table in BIOS. Mostly for MCA use, but
855 * others may find it useful: 824 * others may find it useful:
diff --git a/arch/x86/include/asm/ptrace-abi.h b/arch/x86/include/asm/ptrace-abi.h
index 86723035a515..52b098a6eebb 100644
--- a/arch/x86/include/asm/ptrace-abi.h
+++ b/arch/x86/include/asm/ptrace-abi.h
@@ -82,61 +82,6 @@
82 82
83#ifndef __ASSEMBLY__ 83#ifndef __ASSEMBLY__
84#include <linux/types.h> 84#include <linux/types.h>
85 85#endif
86/* configuration/status structure used in PTRACE_BTS_CONFIG and
87 PTRACE_BTS_STATUS commands.
88*/
89struct ptrace_bts_config {
90 /* requested or actual size of BTS buffer in bytes */
91 __u32 size;
92 /* bitmask of below flags */
93 __u32 flags;
94 /* buffer overflow signal */
95 __u32 signal;
96 /* actual size of bts_struct in bytes */
97 __u32 bts_size;
98};
99#endif /* __ASSEMBLY__ */
100
101#define PTRACE_BTS_O_TRACE 0x1 /* branch trace */
102#define PTRACE_BTS_O_SCHED 0x2 /* scheduling events w/ jiffies */
103#define PTRACE_BTS_O_SIGNAL 0x4 /* send SIG<signal> on buffer overflow
104 instead of wrapping around */
105#define PTRACE_BTS_O_ALLOC 0x8 /* (re)allocate buffer */
106
107#define PTRACE_BTS_CONFIG 40
108/* Configure branch trace recording.
109 ADDR points to a struct ptrace_bts_config.
110 DATA gives the size of that buffer.
111 A new buffer is allocated, if requested in the flags.
112 An overflow signal may only be requested for new buffers.
113 Returns the number of bytes read.
114*/
115#define PTRACE_BTS_STATUS 41
116/* Return the current configuration in a struct ptrace_bts_config
117 pointed to by ADDR; DATA gives the size of that buffer.
118 Returns the number of bytes written.
119*/
120#define PTRACE_BTS_SIZE 42
121/* Return the number of available BTS records for draining.
122 DATA and ADDR are ignored.
123*/
124#define PTRACE_BTS_GET 43
125/* Get a single BTS record.
126 DATA defines the index into the BTS array, where 0 is the newest
127 entry, and higher indices refer to older entries.
128 ADDR is pointing to struct bts_struct (see asm/ds.h).
129*/
130#define PTRACE_BTS_CLEAR 44
131/* Clear the BTS buffer.
132 DATA and ADDR are ignored.
133*/
134#define PTRACE_BTS_DRAIN 45
135/* Read all available BTS records and clear the buffer.
136 ADDR points to an array of struct bts_struct.
137 DATA gives the size of that buffer.
138 BTS records are read from oldest to newest.
139 Returns number of BTS records drained.
140*/
141 86
142#endif /* _ASM_X86_PTRACE_ABI_H */ 87#endif /* _ASM_X86_PTRACE_ABI_H */
diff --git a/arch/x86/include/asm/ptrace.h b/arch/x86/include/asm/ptrace.h
index 69a686a7dff0..78cd1ea94500 100644
--- a/arch/x86/include/asm/ptrace.h
+++ b/arch/x86/include/asm/ptrace.h
@@ -289,12 +289,6 @@ extern int do_get_thread_area(struct task_struct *p, int idx,
289extern int do_set_thread_area(struct task_struct *p, int idx, 289extern int do_set_thread_area(struct task_struct *p, int idx,
290 struct user_desc __user *info, int can_allocate); 290 struct user_desc __user *info, int can_allocate);
291 291
292#ifdef CONFIG_X86_PTRACE_BTS
293extern void ptrace_bts_untrace(struct task_struct *tsk);
294
295#define arch_ptrace_untrace(tsk) ptrace_bts_untrace(tsk)
296#endif /* CONFIG_X86_PTRACE_BTS */
297
298#endif /* __KERNEL__ */ 292#endif /* __KERNEL__ */
299 293
300#endif /* !__ASSEMBLY__ */ 294#endif /* !__ASSEMBLY__ */
diff --git a/arch/x86/include/asm/thread_info.h b/arch/x86/include/asm/thread_info.h
index e0d28901e969..d017ed5502e2 100644
--- a/arch/x86/include/asm/thread_info.h
+++ b/arch/x86/include/asm/thread_info.h
@@ -92,8 +92,7 @@ struct thread_info {
92#define TIF_IO_BITMAP 22 /* uses I/O bitmap */ 92#define TIF_IO_BITMAP 22 /* uses I/O bitmap */
93#define TIF_FREEZE 23 /* is freezing for suspend */ 93#define TIF_FREEZE 23 /* is freezing for suspend */
94#define TIF_FORCED_TF 24 /* true if TF in eflags artificially */ 94#define TIF_FORCED_TF 24 /* true if TF in eflags artificially */
95#define TIF_DEBUGCTLMSR 25 /* uses thread_struct.debugctlmsr */ 95#define TIF_BLOCKSTEP 25 /* set when we want DEBUGCTLMSR_BTF */
96#define TIF_DS_AREA_MSR 26 /* uses thread_struct.ds_area_msr */
97#define TIF_LAZY_MMU_UPDATES 27 /* task is updating the mmu lazily */ 96#define TIF_LAZY_MMU_UPDATES 27 /* task is updating the mmu lazily */
98#define TIF_SYSCALL_TRACEPOINT 28 /* syscall tracepoint instrumentation */ 97#define TIF_SYSCALL_TRACEPOINT 28 /* syscall tracepoint instrumentation */
99 98
@@ -115,8 +114,7 @@ struct thread_info {
115#define _TIF_IO_BITMAP (1 << TIF_IO_BITMAP) 114#define _TIF_IO_BITMAP (1 << TIF_IO_BITMAP)
116#define _TIF_FREEZE (1 << TIF_FREEZE) 115#define _TIF_FREEZE (1 << TIF_FREEZE)
117#define _TIF_FORCED_TF (1 << TIF_FORCED_TF) 116#define _TIF_FORCED_TF (1 << TIF_FORCED_TF)
118#define _TIF_DEBUGCTLMSR (1 << TIF_DEBUGCTLMSR) 117#define _TIF_BLOCKSTEP (1 << TIF_BLOCKSTEP)
119#define _TIF_DS_AREA_MSR (1 << TIF_DS_AREA_MSR)
120#define _TIF_LAZY_MMU_UPDATES (1 << TIF_LAZY_MMU_UPDATES) 118#define _TIF_LAZY_MMU_UPDATES (1 << TIF_LAZY_MMU_UPDATES)
121#define _TIF_SYSCALL_TRACEPOINT (1 << TIF_SYSCALL_TRACEPOINT) 119#define _TIF_SYSCALL_TRACEPOINT (1 << TIF_SYSCALL_TRACEPOINT)
122 120
@@ -147,7 +145,7 @@ struct thread_info {
147 145
148/* flags to check in __switch_to() */ 146/* flags to check in __switch_to() */
149#define _TIF_WORK_CTXSW \ 147#define _TIF_WORK_CTXSW \
150 (_TIF_IO_BITMAP|_TIF_DEBUGCTLMSR|_TIF_DS_AREA_MSR|_TIF_NOTSC) 148 (_TIF_IO_BITMAP|_TIF_NOTSC|_TIF_BLOCKSTEP)
151 149
152#define _TIF_WORK_CTXSW_PREV (_TIF_WORK_CTXSW|_TIF_USER_RETURN_NOTIFY) 150#define _TIF_WORK_CTXSW_PREV (_TIF_WORK_CTXSW|_TIF_USER_RETURN_NOTIFY)
153#define _TIF_WORK_CTXSW_NEXT (_TIF_WORK_CTXSW|_TIF_DEBUG) 151#define _TIF_WORK_CTXSW_NEXT (_TIF_WORK_CTXSW|_TIF_DEBUG)
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-11 20:54:32 -0500