aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--arch/x86/Kconfig.cpu20
-rw-r--r--arch/x86/Kconfig.debug9
-rw-r--r--arch/x86/include/asm/ds.h302
-rw-r--r--arch/x86/include/asm/processor.h33
-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.h6
-rw-r--r--arch/x86/kernel/Makefile2
-rw-r--r--arch/x86/kernel/cpu/intel.c2
-rw-r--r--arch/x86/kernel/ds.c1437
-rw-r--r--arch/x86/kernel/ds_selftest.c408
-rw-r--r--arch/x86/kernel/ds_selftest.h15
-rw-r--r--arch/x86/kernel/dumpstack.c5
-rw-r--r--arch/x86/kernel/kprobes.c6
-rw-r--r--arch/x86/kernel/process.c9
-rw-r--r--arch/x86/kernel/process_32.c8
-rw-r--r--arch/x86/kernel/process_64.c8
-rw-r--r--arch/x86/kernel/ptrace.c382
-rw-r--r--arch/x86/kernel/step.c36
-rw-r--r--arch/x86/kernel/traps.c5
-rw-r--r--include/linux/ftrace.h12
-rw-r--r--include/linux/mm.h4
-rw-r--r--include/linux/ptrace.h12
-rw-r--r--include/linux/sched.h9
-rw-r--r--kernel/fork.c3
-rw-r--r--kernel/ptrace.c1
-rw-r--r--kernel/sched.c43
-rw-r--r--kernel/trace/Kconfig11
-rw-r--r--kernel/trace/Makefile1
-rw-r--r--kernel/trace/trace.h4
-rw-r--r--kernel/trace/trace_entries.h12
-rw-r--r--kernel/trace/trace_hw_branches.c312
-rw-r--r--kernel/trace/trace_selftest.c57
-rw-r--r--mm/mlock.c41
34 files changed, 9 insertions, 3269 deletions
diff --git a/arch/x86/Kconfig.cpu b/arch/x86/Kconfig.cpu
index a19829374e6a..918fbb1855cc 100644
--- a/arch/x86/Kconfig.cpu
+++ b/arch/x86/Kconfig.cpu
@@ -502,23 +502,3 @@ config CPU_SUP_UMC_32
502 CPU might render the kernel unbootable. 502 CPU might render the kernel unbootable.
503 503
504 If unsure, say N. 504 If unsure, say N.
505
506config X86_DS
507 def_bool X86_PTRACE_BTS
508 depends on X86_DEBUGCTLMSR
509 select HAVE_HW_BRANCH_TRACER
510
511config X86_PTRACE_BTS
512 bool "Branch Trace Store"
513 default y
514 depends on X86_DEBUGCTLMSR
515 depends on BROKEN
516 ---help---
517 This adds a ptrace interface to the hardware's branch trace store.
518
519 Debuggers may use it to collect an execution trace of the debugged
520 application in order to answer the question 'how did I get here?'.
521 Debuggers may trace user mode as well as kernel mode.
522
523 Say Y unless there is no application development on this machine
524 and you want to save a small amount of code size.
diff --git a/arch/x86/Kconfig.debug b/arch/x86/Kconfig.debug
index bc01e3ebfeb2..bd58c8abbfbd 100644
--- a/arch/x86/Kconfig.debug
+++ b/arch/x86/Kconfig.debug
@@ -174,15 +174,6 @@ config IOMMU_LEAK
174 Add a simple leak tracer to the IOMMU code. This is useful when you 174 Add a simple leak tracer to the IOMMU code. This is useful when you
175 are debugging a buggy device driver that leaks IOMMU mappings. 175 are debugging a buggy device driver that leaks IOMMU mappings.
176 176
177config X86_DS_SELFTEST
178 bool "DS selftest"
179 default y
180 depends on DEBUG_KERNEL
181 depends on X86_DS
182 ---help---
183 Perform Debug Store selftests at boot time.
184 If in doubt, say "N".
185
186config HAVE_MMIOTRACE_SUPPORT 177config HAVE_MMIOTRACE_SUPPORT
187 def_bool y 178 def_bool y
188 179
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/processor.h b/arch/x86/include/asm/processor.h
index b753ea59703a..5bec21a66dc5 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)
@@ -814,21 +810,6 @@ static inline unsigned long get_debugctlmsr(void)
814 return debugctlmsr; 810 return debugctlmsr;
815} 811}
816 812
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;
830}
831
832static inline void update_debugctlmsr(unsigned long debugctlmsr) 813static inline void update_debugctlmsr(unsigned long debugctlmsr)
833{ 814{
834#ifndef CONFIG_X86_DEBUGCTLMSR 815#ifndef CONFIG_X86_DEBUGCTLMSR
@@ -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..dc85e12d1405 100644
--- a/arch/x86/include/asm/thread_info.h
+++ b/arch/x86/include/asm/thread_info.h
@@ -92,8 +92,6 @@ 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 */
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 */ 95#define TIF_LAZY_MMU_UPDATES 27 /* task is updating the mmu lazily */
98#define TIF_SYSCALL_TRACEPOINT 28 /* syscall tracepoint instrumentation */ 96#define TIF_SYSCALL_TRACEPOINT 28 /* syscall tracepoint instrumentation */
99 97
@@ -115,8 +113,6 @@ struct thread_info {
115#define _TIF_IO_BITMAP (1 << TIF_IO_BITMAP) 113#define _TIF_IO_BITMAP (1 << TIF_IO_BITMAP)
116#define _TIF_FREEZE (1 << TIF_FREEZE) 114#define _TIF_FREEZE (1 << TIF_FREEZE)
117#define _TIF_FORCED_TF (1 << TIF_FORCED_TF) 115#define _TIF_FORCED_TF (1 << TIF_FORCED_TF)
118#define _TIF_DEBUGCTLMSR (1 << TIF_DEBUGCTLMSR)
119#define _TIF_DS_AREA_MSR (1 << TIF_DS_AREA_MSR)
120#define _TIF_LAZY_MMU_UPDATES (1 << TIF_LAZY_MMU_UPDATES) 116#define _TIF_LAZY_MMU_UPDATES (1 << TIF_LAZY_MMU_UPDATES)
121#define _TIF_SYSCALL_TRACEPOINT (1 << TIF_SYSCALL_TRACEPOINT) 117#define _TIF_SYSCALL_TRACEPOINT (1 << TIF_SYSCALL_TRACEPOINT)
122 118
@@ -147,7 +143,7 @@ struct thread_info {
147 143
148/* flags to check in __switch_to() */ 144/* flags to check in __switch_to() */
149#define _TIF_WORK_CTXSW \ 145#define _TIF_WORK_CTXSW \
150 (_TIF_IO_BITMAP|_TIF_DEBUGCTLMSR|_TIF_DS_AREA_MSR|_TIF_NOTSC) 146 (_TIF_IO_BITMAP|_TIF_NOTSC)
151 147
152#define _TIF_WORK_CTXSW_PREV (_TIF_WORK_CTXSW|_TIF_USER_RETURN_NOTIFY) 148#define _TIF_WORK_CTXSW_PREV (_TIF_WORK_CTXSW|_TIF_USER_RETURN_NOTIFY)
153#define _TIF_WORK_CTXSW_NEXT (_TIF_WORK_CTXSW|_TIF_DEBUG) 149#define _TIF_WORK_CTXSW_NEXT (_TIF_WORK_CTXSW|_TIF_DEBUG)
diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile
index 4c58352209e0..e77b22083721 100644
--- a/arch/x86/kernel/Makefile
+++ b/arch/x86/kernel/Makefile
@@ -47,8 +47,6 @@ obj-$(CONFIG_X86_TRAMPOLINE) += trampoline.o
47obj-y += process.o 47obj-y += process.o
48obj-y += i387.o xsave.o 48obj-y += i387.o xsave.o
49obj-y += ptrace.o 49obj-y += ptrace.o
50obj-$(CONFIG_X86_DS) += ds.o
51obj-$(CONFIG_X86_DS_SELFTEST) += ds_selftest.o
52obj-$(CONFIG_X86_32) += tls.o 50obj-$(CONFIG_X86_32) += tls.o
53obj-$(CONFIG_IA32_EMULATION) += tls.o 51obj-$(CONFIG_IA32_EMULATION) += tls.o
54obj-y += step.o 52obj-y += step.o
diff --git a/arch/x86/kernel/cpu/intel.c b/arch/x86/kernel/cpu/intel.c
index 7e1cca13af35..d72377c41c76 100644
--- a/arch/x86/kernel/cpu/intel.c
+++ b/arch/x86/kernel/cpu/intel.c
@@ -12,7 +12,6 @@
12#include <asm/processor.h> 12#include <asm/processor.h>
13#include <asm/pgtable.h> 13#include <asm/pgtable.h>
14#include <asm/msr.h> 14#include <asm/msr.h>
15#include <asm/ds.h>
16#include <asm/bugs.h> 15#include <asm/bugs.h>
17#include <asm/cpu.h> 16#include <asm/cpu.h>
18 17
@@ -367,7 +366,6 @@ static void __cpuinit init_intel(struct cpuinfo_x86 *c)
367 set_cpu_cap(c, X86_FEATURE_BTS); 366 set_cpu_cap(c, X86_FEATURE_BTS);
368 if (!(l1 & (1<<12))) 367 if (!(l1 & (1<<12)))
369 set_cpu_cap(c, X86_FEATURE_PEBS); 368 set_cpu_cap(c, X86_FEATURE_PEBS);
370 ds_init_intel(c);
371 } 369 }
372 370
373 if (c->x86 == 6 && c->x86_model == 29 && cpu_has_clflush) 371 if (c->x86 == 6 && c->x86_model == 29 && cpu_has_clflush)
diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c
deleted file mode 100644
index 1c47390dd0e5..000000000000
--- a/arch/x86/kernel/ds.c
+++ /dev/null
@@ -1,1437 +0,0 @@
1/*
2 * Debug Store 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#include <linux/kernel.h>
23#include <linux/string.h>
24#include <linux/errno.h>
25#include <linux/sched.h>
26#include <linux/slab.h>
27#include <linux/mm.h>
28#include <linux/trace_clock.h>
29
30#include <asm/ds.h>
31
32#include "ds_selftest.h"
33
34/*
35 * The configuration for a particular DS hardware implementation:
36 */
37struct ds_configuration {
38 /* The name of the configuration: */
39 const char *name;
40
41 /* The size of pointer-typed fields in DS, BTS, and PEBS: */
42 unsigned char sizeof_ptr_field;
43
44 /* The size of a BTS/PEBS record in bytes: */
45 unsigned char sizeof_rec[2];
46
47 /* The number of pebs counter reset values in the DS structure. */
48 unsigned char nr_counter_reset;
49
50 /* Control bit-masks indexed by enum ds_feature: */
51 unsigned long ctl[dsf_ctl_max];
52};
53static struct ds_configuration ds_cfg __read_mostly;
54
55
56/* Maximal size of a DS configuration: */
57#define MAX_SIZEOF_DS 0x80
58
59/* Maximal size of a BTS record: */
60#define MAX_SIZEOF_BTS (3 * 8)
61
62/* BTS and PEBS buffer alignment: */
63#define DS_ALIGNMENT (1 << 3)
64
65/* Number of buffer pointers in DS: */
66#define NUM_DS_PTR_FIELDS 8
67
68/* Size of a pebs reset value in DS: */
69#define PEBS_RESET_FIELD_SIZE 8
70
71/* Mask of control bits in the DS MSR register: */
72#define BTS_CONTROL \
73 ( ds_cfg.ctl[dsf_bts] | \
74 ds_cfg.ctl[dsf_bts_kernel] | \
75 ds_cfg.ctl[dsf_bts_user] | \
76 ds_cfg.ctl[dsf_bts_overflow] )
77
78/*
79 * A BTS or PEBS tracer.
80 *
81 * This holds the configuration of the tracer and serves as a handle
82 * to identify tracers.
83 */
84struct ds_tracer {
85 /* The DS context (partially) owned by this tracer. */
86 struct ds_context *context;
87 /* The buffer provided on ds_request() and its size in bytes. */
88 void *buffer;
89 size_t size;
90};
91
92struct bts_tracer {
93 /* The common DS part: */
94 struct ds_tracer ds;
95
96 /* The trace including the DS configuration: */
97 struct bts_trace trace;
98
99 /* Buffer overflow notification function: */
100 bts_ovfl_callback_t ovfl;
101
102 /* Active flags affecting trace collection. */
103 unsigned int flags;
104};
105
106struct pebs_tracer {
107 /* The common DS part: */
108 struct ds_tracer ds;
109
110 /* The trace including the DS configuration: */
111 struct pebs_trace trace;
112
113 /* Buffer overflow notification function: */
114 pebs_ovfl_callback_t ovfl;
115};
116
117/*
118 * Debug Store (DS) save area configuration (see Intel64 and IA32
119 * Architectures Software Developer's Manual, section 18.5)
120 *
121 * The DS configuration consists of the following fields; different
122 * architetures vary in the size of those fields.
123 *
124 * - double-word aligned base linear address of the BTS buffer
125 * - write pointer into the BTS buffer
126 * - end linear address of the BTS buffer (one byte beyond the end of
127 * the buffer)
128 * - interrupt pointer into BTS buffer
129 * (interrupt occurs when write pointer passes interrupt pointer)
130 * - double-word aligned base linear address of the PEBS buffer
131 * - write pointer into the PEBS buffer
132 * - end linear address of the PEBS buffer (one byte beyond the end of
133 * the buffer)
134 * - interrupt pointer into PEBS buffer
135 * (interrupt occurs when write pointer passes interrupt pointer)
136 * - value to which counter is reset following counter overflow
137 *
138 * Later architectures use 64bit pointers throughout, whereas earlier
139 * architectures use 32bit pointers in 32bit mode.
140 *
141 *
142 * We compute the base address for the first 8 fields based on:
143 * - the field size stored in the DS configuration
144 * - the relative field position
145 * - an offset giving the start of the respective region
146 *
147 * This offset is further used to index various arrays holding
148 * information for BTS and PEBS at the respective index.
149 *
150 * On later 32bit processors, we only access the lower 32bit of the
151 * 64bit pointer fields. The upper halves will be zeroed out.
152 */
153
154enum ds_field {
155 ds_buffer_base = 0,
156 ds_index,
157 ds_absolute_maximum,
158 ds_interrupt_threshold,
159};
160
161enum ds_qualifier {
162 ds_bts = 0,
163 ds_pebs
164};
165
166static inline unsigned long
167ds_get(const unsigned char *base, enum ds_qualifier qual, enum ds_field field)
168{
169 base += (ds_cfg.sizeof_ptr_field * (field + (4 * qual)));
170 return *(unsigned long *)base;
171}
172
173static inline void
174ds_set(unsigned char *base, enum ds_qualifier qual, enum ds_field field,
175 unsigned long value)
176{
177 base += (ds_cfg.sizeof_ptr_field * (field + (4 * qual)));
178 (*(unsigned long *)base) = value;
179}
180
181
182/*
183 * Locking is done only for allocating BTS or PEBS resources.
184 */
185static DEFINE_SPINLOCK(ds_lock);
186
187/*
188 * We either support (system-wide) per-cpu or per-thread allocation.
189 * We distinguish the two based on the task_struct pointer, where a
190 * NULL pointer indicates per-cpu allocation for the current cpu.
191 *
192 * Allocations are use-counted. As soon as resources are allocated,
193 * further allocations must be of the same type (per-cpu or
194 * per-thread). We model this by counting allocations (i.e. the number
195 * of tracers of a certain type) for one type negatively:
196 * =0 no tracers
197 * >0 number of per-thread tracers
198 * <0 number of per-cpu tracers
199 *
200 * Tracers essentially gives the number of ds contexts for a certain
201 * type of allocation.
202 */
203static atomic_t tracers = ATOMIC_INIT(0);
204
205static inline int get_tracer(struct task_struct *task)
206{
207 int error;
208
209 spin_lock_irq(&ds_lock);
210
211 if (task) {
212 error = -EPERM;
213 if (atomic_read(&tracers) < 0)
214 goto out;
215 atomic_inc(&tracers);
216 } else {
217 error = -EPERM;
218 if (atomic_read(&tracers) > 0)
219 goto out;
220 atomic_dec(&tracers);
221 }
222
223 error = 0;
224out:
225 spin_unlock_irq(&ds_lock);
226 return error;
227}
228
229static inline void put_tracer(struct task_struct *task)
230{
231 if (task)
232 atomic_dec(&tracers);
233 else
234 atomic_inc(&tracers);
235}
236
237/*
238 * The DS context is either attached to a thread or to a cpu:
239 * - in the former case, the thread_struct contains a pointer to the
240 * attached context.
241 * - in the latter case, we use a static array of per-cpu context
242 * pointers.
243 *
244 * Contexts are use-counted. They are allocated on first access and
245 * deallocated when the last user puts the context.
246 */
247struct ds_context {
248 /* The DS configuration; goes into MSR_IA32_DS_AREA: */
249 unsigned char ds[MAX_SIZEOF_DS];
250
251 /* The owner of the BTS and PEBS configuration, respectively: */
252 struct bts_tracer *bts_master;
253 struct pebs_tracer *pebs_master;
254
255 /* Use count: */
256 unsigned long count;
257
258 /* Pointer to the context pointer field: */
259 struct ds_context **this;
260
261 /* The traced task; NULL for cpu tracing: */
262 struct task_struct *task;
263
264 /* The traced cpu; only valid if task is NULL: */
265 int cpu;
266};
267
268static DEFINE_PER_CPU(struct ds_context *, cpu_ds_context);
269
270
271static struct ds_context *ds_get_context(struct task_struct *task, int cpu)
272{
273 struct ds_context **p_context =
274 (task ? &task->thread.ds_ctx : &per_cpu(cpu_ds_context, cpu));
275 struct ds_context *context = NULL;
276 struct ds_context *new_context = NULL;
277
278 /* Chances are small that we already have a context. */
279 new_context = kzalloc(sizeof(*new_context), GFP_KERNEL);
280 if (!new_context)
281 return NULL;
282
283 spin_lock_irq(&ds_lock);
284
285 context = *p_context;
286 if (likely(!context)) {
287 context = new_context;
288
289 context->this = p_context;
290 context->task = task;
291 context->cpu = cpu;
292 context->count = 0;
293
294 *p_context = context;
295 }
296
297 context->count++;
298
299 spin_unlock_irq(&ds_lock);
300
301 if (context != new_context)
302 kfree(new_context);
303
304 return context;
305}
306
307static void ds_put_context(struct ds_context *context)
308{
309 struct task_struct *task;
310 unsigned long irq;
311
312 if (!context)
313 return;
314
315 spin_lock_irqsave(&ds_lock, irq);
316
317 if (--context->count) {
318 spin_unlock_irqrestore(&ds_lock, irq);
319 return;
320 }
321
322 *(context->this) = NULL;
323
324 task = context->task;
325
326 if (task)
327 clear_tsk_thread_flag(task, TIF_DS_AREA_MSR);
328
329 /*
330 * We leave the (now dangling) pointer to the DS configuration in
331 * the DS_AREA msr. This is as good or as bad as replacing it with
332 * NULL - the hardware would crash if we enabled tracing.
333 *
334 * This saves us some problems with having to write an msr on a
335 * different cpu while preventing others from doing the same for the
336 * next context for that same cpu.
337 */
338
339 spin_unlock_irqrestore(&ds_lock, irq);
340
341 /* The context might still be in use for context switching. */
342 if (task && (task != current))
343 wait_task_context_switch(task);
344
345 kfree(context);
346}
347
348static void ds_install_ds_area(struct ds_context *context)
349{
350 unsigned long ds;
351
352 ds = (unsigned long)context->ds;
353
354 /*
355 * There is a race between the bts master and the pebs master.
356 *
357 * The thread/cpu access is synchronized via get/put_cpu() for
358 * task tracing and via wrmsr_on_cpu for cpu tracing.
359 *
360 * If bts and pebs are collected for the same task or same cpu,
361 * the same confiuration is written twice.
362 */
363 if (context->task) {
364 get_cpu();
365 if (context->task == current)
366 wrmsrl(MSR_IA32_DS_AREA, ds);
367 set_tsk_thread_flag(context->task, TIF_DS_AREA_MSR);
368 put_cpu();
369 } else
370 wrmsr_on_cpu(context->cpu, MSR_IA32_DS_AREA,
371 (u32)((u64)ds), (u32)((u64)ds >> 32));
372}
373
374/*
375 * Call the tracer's callback on a buffer overflow.
376 *
377 * context: the ds context
378 * qual: the buffer type
379 */
380static void ds_overflow(struct ds_context *context, enum ds_qualifier qual)
381{
382 switch (qual) {
383 case ds_bts:
384 if (context->bts_master &&
385 context->bts_master->ovfl)
386 context->bts_master->ovfl(context->bts_master);
387 break;
388 case ds_pebs:
389 if (context->pebs_master &&
390 context->pebs_master->ovfl)
391 context->pebs_master->ovfl(context->pebs_master);
392 break;
393 }
394}
395
396
397/*
398 * Write raw data into the BTS or PEBS buffer.
399 *
400 * The remainder of any partially written record is zeroed out.
401 *
402 * context: the DS context
403 * qual: the buffer type
404 * record: the data to write
405 * size: the size of the data
406 */
407static int ds_write(struct ds_context *context, enum ds_qualifier qual,
408 const void *record, size_t size)
409{
410 int bytes_written = 0;
411
412 if (!record)
413 return -EINVAL;
414
415 while (size) {
416 unsigned long base, index, end, write_end, int_th;
417 unsigned long write_size, adj_write_size;
418
419 /*
420 * Write as much as possible without producing an
421 * overflow interrupt.
422 *
423 * Interrupt_threshold must either be
424 * - bigger than absolute_maximum or
425 * - point to a record between buffer_base and absolute_maximum
426 *
427 * Index points to a valid record.
428 */
429 base = ds_get(context->ds, qual, ds_buffer_base);
430 index = ds_get(context->ds, qual, ds_index);
431 end = ds_get(context->ds, qual, ds_absolute_maximum);
432 int_th = ds_get(context->ds, qual, ds_interrupt_threshold);
433
434 write_end = min(end, int_th);
435
436 /*
437 * If we are already beyond the interrupt threshold,
438 * we fill the entire buffer.
439 */
440 if (write_end <= index)
441 write_end = end;
442
443 if (write_end <= index)
444 break;
445
446 write_size = min((unsigned long) size, write_end - index);
447 memcpy((void *)index, record, write_size);
448
449 record = (const char *)record + write_size;
450 size -= write_size;
451 bytes_written += write_size;
452
453 adj_write_size = write_size / ds_cfg.sizeof_rec[qual];
454 adj_write_size *= ds_cfg.sizeof_rec[qual];
455
456 /* Zero out trailing bytes. */
457 memset((char *)index + write_size, 0,
458 adj_write_size - write_size);
459 index += adj_write_size;
460
461 if (index >= end)
462 index = base;
463 ds_set(context->ds, qual, ds_index, index);
464
465 if (index >= int_th)
466 ds_overflow(context, qual);
467 }
468
469 return bytes_written;
470}
471
472
473/*
474 * Branch Trace Store (BTS) uses the following format. Different
475 * architectures vary in the size of those fields.
476 * - source linear address
477 * - destination linear address
478 * - flags
479 *
480 * Later architectures use 64bit pointers throughout, whereas earlier
481 * architectures use 32bit pointers in 32bit mode.
482 *
483 * We compute the base address for the fields based on:
484 * - the field size stored in the DS configuration
485 * - the relative field position
486 *
487 * In order to store additional information in the BTS buffer, we use
488 * a special source address to indicate that the record requires
489 * special interpretation.
490 *
491 * Netburst indicated via a bit in the flags field whether the branch
492 * was predicted; this is ignored.
493 *
494 * We use two levels of abstraction:
495 * - the raw data level defined here
496 * - an arch-independent level defined in ds.h
497 */
498
499enum bts_field {
500 bts_from,
501 bts_to,
502 bts_flags,
503
504 bts_qual = bts_from,
505 bts_clock = bts_to,
506 bts_pid = bts_flags,
507
508 bts_qual_mask = (bts_qual_max - 1),
509 bts_escape = ((unsigned long)-1 & ~bts_qual_mask)
510};
511
512static inline unsigned long bts_get(const char *base, unsigned long field)
513{
514 base += (ds_cfg.sizeof_ptr_field * field);
515 return *(unsigned long *)base;
516}
517
518static inline void bts_set(char *base, unsigned long field, unsigned long val)
519{
520 base += (ds_cfg.sizeof_ptr_field * field);
521 (*(unsigned long *)base) = val;
522}
523
524
525/*
526 * The raw BTS data is architecture dependent.
527 *
528 * For higher-level users, we give an arch-independent view.
529 * - ds.h defines struct bts_struct
530 * - bts_read translates one raw bts record into a bts_struct
531 * - bts_write translates one bts_struct into the raw format and
532 * writes it into the top of the parameter tracer's buffer.
533 *
534 * return: bytes read/written on success; -Eerrno, otherwise
535 */
536static int
537bts_read(struct bts_tracer *tracer, const void *at, struct bts_struct *out)
538{
539 if (!tracer)
540 return -EINVAL;
541
542 if (at < tracer->trace.ds.begin)
543 return -EINVAL;
544
545 if (tracer->trace.ds.end < (at + tracer->trace.ds.size))
546 return -EINVAL;
547
548 memset(out, 0, sizeof(*out));
549 if ((bts_get(at, bts_qual) & ~bts_qual_mask) == bts_escape) {
550 out->qualifier = (bts_get(at, bts_qual) & bts_qual_mask);
551 out->variant.event.clock = bts_get(at, bts_clock);
552 out->variant.event.pid = bts_get(at, bts_pid);
553 } else {
554 out->qualifier = bts_branch;
555 out->variant.lbr.from = bts_get(at, bts_from);
556 out->variant.lbr.to = bts_get(at, bts_to);
557
558 if (!out->variant.lbr.from && !out->variant.lbr.to)
559 out->qualifier = bts_invalid;
560 }
561
562 return ds_cfg.sizeof_rec[ds_bts];
563}
564
565static int bts_write(struct bts_tracer *tracer, const struct bts_struct *in)
566{
567 unsigned char raw[MAX_SIZEOF_BTS];
568
569 if (!tracer)
570 return -EINVAL;
571
572 if (MAX_SIZEOF_BTS < ds_cfg.sizeof_rec[ds_bts])
573 return -EOVERFLOW;
574
575 switch (in->qualifier) {
576 case bts_invalid:
577 bts_set(raw, bts_from, 0);
578 bts_set(raw, bts_to, 0);
579 bts_set(raw, bts_flags, 0);
580 break;
581 case bts_branch:
582 bts_set(raw, bts_from, in->variant.lbr.from);
583 bts_set(raw, bts_to, in->variant.lbr.to);
584 bts_set(raw, bts_flags, 0);
585 break;
586 case bts_task_arrives:
587 case bts_task_departs:
588 bts_set(raw, bts_qual, (bts_escape | in->qualifier));
589 bts_set(raw, bts_clock, in->variant.event.clock);
590 bts_set(raw, bts_pid, in->variant.event.pid);
591 break;
592 default:
593 return -EINVAL;
594 }
595
596 return ds_write(tracer->ds.context, ds_bts, raw,
597 ds_cfg.sizeof_rec[ds_bts]);
598}
599
600
601static void ds_write_config(struct ds_context *context,
602 struct ds_trace *cfg, enum ds_qualifier qual)
603{
604 unsigned char *ds = context->ds;
605
606 ds_set(ds, qual, ds_buffer_base, (unsigned long)cfg->begin);
607 ds_set(ds, qual, ds_index, (unsigned long)cfg->top);
608 ds_set(ds, qual, ds_absolute_maximum, (unsigned long)cfg->end);
609 ds_set(ds, qual, ds_interrupt_threshold, (unsigned long)cfg->ith);
610}
611
612static void ds_read_config(struct ds_context *context,
613 struct ds_trace *cfg, enum ds_qualifier qual)
614{
615 unsigned char *ds = context->ds;
616
617 cfg->begin = (void *)ds_get(ds, qual, ds_buffer_base);
618 cfg->top = (void *)ds_get(ds, qual, ds_index);
619 cfg->end = (void *)ds_get(ds, qual, ds_absolute_maximum);
620 cfg->ith = (void *)ds_get(ds, qual, ds_interrupt_threshold);
621}
622
623static void ds_init_ds_trace(struct ds_trace *trace, enum ds_qualifier qual,
624 void *base, size_t size, size_t ith,
625 unsigned int flags) {
626 unsigned long buffer, adj;
627
628 /*
629 * Adjust the buffer address and size to meet alignment
630 * constraints:
631 * - buffer is double-word aligned
632 * - size is multiple of record size
633 *
634 * We checked the size at the very beginning; we have enough
635 * space to do the adjustment.
636 */
637 buffer = (unsigned long)base;
638
639 adj = ALIGN(buffer, DS_ALIGNMENT) - buffer;
640 buffer += adj;
641 size -= adj;
642
643 trace->n = size / ds_cfg.sizeof_rec[qual];
644 trace->size = ds_cfg.sizeof_rec[qual];
645
646 size = (trace->n * trace->size);
647
648 trace->begin = (void *)buffer;
649 trace->top = trace->begin;
650 trace->end = (void *)(buffer + size);
651 /*
652 * The value for 'no threshold' is -1, which will set the
653 * threshold outside of the buffer, just like we want it.
654 */
655 ith *= ds_cfg.sizeof_rec[qual];
656 trace->ith = (void *)(buffer + size - ith);
657
658 trace->flags = flags;
659}
660
661
662static int ds_request(struct ds_tracer *tracer, struct ds_trace *trace,
663 enum ds_qualifier qual, struct task_struct *task,
664 int cpu, void *base, size_t size, size_t th)
665{
666 struct ds_context *context;
667 int error;
668 size_t req_size;
669
670 error = -EOPNOTSUPP;
671 if (!ds_cfg.sizeof_rec[qual])
672 goto out;
673
674 error = -EINVAL;
675 if (!base)
676 goto out;
677
678 req_size = ds_cfg.sizeof_rec[qual];
679 /* We might need space for alignment adjustments. */
680 if (!IS_ALIGNED((unsigned long)base, DS_ALIGNMENT))
681 req_size += DS_ALIGNMENT;
682
683 error = -EINVAL;
684 if (size < req_size)
685 goto out;
686
687 if (th != (size_t)-1) {
688 th *= ds_cfg.sizeof_rec[qual];
689
690 error = -EINVAL;
691 if (size <= th)
692 goto out;
693 }
694
695 tracer->buffer = base;
696 tracer->size = size;
697
698 error = -ENOMEM;
699 context = ds_get_context(task, cpu);
700 if (!context)
701 goto out;
702 tracer->context = context;
703
704 /*
705 * Defer any tracer-specific initialization work for the context until
706 * context ownership has been clarified.
707 */
708
709 error = 0;
710 out:
711 return error;
712}
713
714static struct bts_tracer *ds_request_bts(struct task_struct *task, int cpu,
715 void *base, size_t size,
716 bts_ovfl_callback_t ovfl, size_t th,
717 unsigned int flags)
718{
719 struct bts_tracer *tracer;
720 int error;
721
722 /* Buffer overflow notification is not yet implemented. */
723 error = -EOPNOTSUPP;
724 if (ovfl)
725 goto out;
726
727 error = get_tracer(task);
728 if (error < 0)
729 goto out;
730
731 error = -ENOMEM;
732 tracer = kzalloc(sizeof(*tracer), GFP_KERNEL);
733 if (!tracer)
734 goto out_put_tracer;
735 tracer->ovfl = ovfl;
736
737 /* Do some more error checking and acquire a tracing context. */
738 error = ds_request(&tracer->ds, &tracer->trace.ds,
739 ds_bts, task, cpu, base, size, th);
740 if (error < 0)
741 goto out_tracer;
742
743 /* Claim the bts part of the tracing context we acquired above. */
744 spin_lock_irq(&ds_lock);
745
746 error = -EPERM;
747 if (tracer->ds.context->bts_master)
748 goto out_unlock;
749 tracer->ds.context->bts_master = tracer;
750
751 spin_unlock_irq(&ds_lock);
752
753 /*
754 * Now that we own the bts part of the context, let's complete the
755 * initialization for that part.
756 */
757 ds_init_ds_trace(&tracer->trace.ds, ds_bts, base, size, th, flags);
758 ds_write_config(tracer->ds.context, &tracer->trace.ds, ds_bts);
759 ds_install_ds_area(tracer->ds.context);
760
761 tracer->trace.read = bts_read;
762 tracer->trace.write = bts_write;
763
764 /* Start tracing. */
765 ds_resume_bts(tracer);
766
767 return tracer;
768
769 out_unlock:
770 spin_unlock_irq(&ds_lock);
771 ds_put_context(tracer->ds.context);
772 out_tracer:
773 kfree(tracer);
774 out_put_tracer:
775 put_tracer(task);
776 out:
777 return ERR_PTR(error);
778}
779
780struct bts_tracer *ds_request_bts_task(struct task_struct *task,
781 void *base, size_t size,
782 bts_ovfl_callback_t ovfl,
783 size_t th, unsigned int flags)
784{
785 return ds_request_bts(task, 0, base, size, ovfl, th, flags);
786}
787
788struct bts_tracer *ds_request_bts_cpu(int cpu, void *base, size_t size,
789 bts_ovfl_callback_t ovfl,
790 size_t th, unsigned int flags)
791{
792 return ds_request_bts(NULL, cpu, base, size, ovfl, th, flags);
793}
794
795static struct pebs_tracer *ds_request_pebs(struct task_struct *task, int cpu,
796 void *base, size_t size,
797 pebs_ovfl_callback_t ovfl, size_t th,
798 unsigned int flags)
799{
800 struct pebs_tracer *tracer;
801 int error;
802
803 /* Buffer overflow notification is not yet implemented. */
804 error = -EOPNOTSUPP;
805 if (ovfl)
806 goto out;
807
808 error = get_tracer(task);
809 if (error < 0)
810 goto out;
811
812 error = -ENOMEM;
813 tracer = kzalloc(sizeof(*tracer), GFP_KERNEL);
814 if (!tracer)
815 goto out_put_tracer;
816 tracer->ovfl = ovfl;
817
818 /* Do some more error checking and acquire a tracing context. */
819 error = ds_request(&tracer->ds, &tracer->trace.ds,
820 ds_pebs, task, cpu, base, size, th);
821 if (error < 0)
822 goto out_tracer;
823
824 /* Claim the pebs part of the tracing context we acquired above. */
825 spin_lock_irq(&ds_lock);
826
827 error = -EPERM;
828 if (tracer->ds.context->pebs_master)
829 goto out_unlock;
830 tracer->ds.context->pebs_master = tracer;
831
832 spin_unlock_irq(&ds_lock);
833
834 /*
835 * Now that we own the pebs part of the context, let's complete the
836 * initialization for that part.
837 */
838 ds_init_ds_trace(&tracer->trace.ds, ds_pebs, base, size, th, flags);
839 ds_write_config(tracer->ds.context, &tracer->trace.ds, ds_pebs);
840 ds_install_ds_area(tracer->ds.context);
841
842 /* Start tracing. */
843 ds_resume_pebs(tracer);
844
845 return tracer;
846
847 out_unlock:
848 spin_unlock_irq(&ds_lock);
849 ds_put_context(tracer->ds.context);
850 out_tracer:
851 kfree(tracer);
852 out_put_tracer:
853 put_tracer(task);
854 out:
855 return ERR_PTR(error);
856}
857
858struct pebs_tracer *ds_request_pebs_task(struct task_struct *task,
859 void *base, size_t size,
860 pebs_ovfl_callback_t ovfl,
861 size_t th, unsigned int flags)
862{
863 return ds_request_pebs(task, 0, base, size, ovfl, th, flags);
864}
865
866struct pebs_tracer *ds_request_pebs_cpu(int cpu, void *base, size_t size,
867 pebs_ovfl_callback_t ovfl,
868 size_t th, unsigned int flags)
869{
870 return ds_request_pebs(NULL, cpu, base, size, ovfl, th, flags);
871}
872
873static void ds_free_bts(struct bts_tracer *tracer)
874{
875 struct task_struct *task;
876
877 task = tracer->ds.context->task;
878
879 WARN_ON_ONCE(tracer->ds.context->bts_master != tracer);
880 tracer->ds.context->bts_master = NULL;
881
882 /* Make sure tracing stopped and the tracer is not in use. */
883 if (task && (task != current))
884 wait_task_context_switch(task);
885
886 ds_put_context(tracer->ds.context);
887 put_tracer(task);
888
889 kfree(tracer);
890}
891
892void ds_release_bts(struct bts_tracer *tracer)
893{
894 might_sleep();
895
896 if (!tracer)
897 return;
898
899 ds_suspend_bts(tracer);
900 ds_free_bts(tracer);
901}
902
903int ds_release_bts_noirq(struct bts_tracer *tracer)
904{
905 struct task_struct *task;
906 unsigned long irq;
907 int error;
908
909 if (!tracer)
910 return 0;
911
912 task = tracer->ds.context->task;
913
914 local_irq_save(irq);
915
916 error = -EPERM;
917 if (!task &&
918 (tracer->ds.context->cpu != smp_processor_id()))
919 goto out;
920
921 error = -EPERM;
922 if (task && (task != current))
923 goto out;
924
925 ds_suspend_bts_noirq(tracer);
926 ds_free_bts(tracer);
927
928 error = 0;
929 out:
930 local_irq_restore(irq);
931 return error;
932}
933
934static void update_task_debugctlmsr(struct task_struct *task,
935 unsigned long debugctlmsr)
936{
937 task->thread.debugctlmsr = debugctlmsr;
938
939 get_cpu();
940 if (task == current)
941 update_debugctlmsr(debugctlmsr);
942 put_cpu();
943}
944
945void ds_suspend_bts(struct bts_tracer *tracer)
946{
947 struct task_struct *task;
948 unsigned long debugctlmsr;
949 int cpu;
950
951 if (!tracer)
952 return;
953
954 tracer->flags = 0;
955
956 task = tracer->ds.context->task;
957 cpu = tracer->ds.context->cpu;
958
959 WARN_ON(!task && irqs_disabled());
960
961 debugctlmsr = (task ?
962 task->thread.debugctlmsr :
963 get_debugctlmsr_on_cpu(cpu));
964 debugctlmsr &= ~BTS_CONTROL;
965
966 if (task)
967 update_task_debugctlmsr(task, debugctlmsr);
968 else
969 update_debugctlmsr_on_cpu(cpu, debugctlmsr);
970}
971
972int ds_suspend_bts_noirq(struct bts_tracer *tracer)
973{
974 struct task_struct *task;
975 unsigned long debugctlmsr, irq;
976 int cpu, error = 0;
977
978 if (!tracer)
979 return 0;
980
981 tracer->flags = 0;
982
983 task = tracer->ds.context->task;
984 cpu = tracer->ds.context->cpu;
985
986 local_irq_save(irq);
987
988 error = -EPERM;
989 if (!task && (cpu != smp_processor_id()))
990 goto out;
991
992 debugctlmsr = (task ?
993 task->thread.debugctlmsr :
994 get_debugctlmsr());
995 debugctlmsr &= ~BTS_CONTROL;
996
997 if (task)
998 update_task_debugctlmsr(task, debugctlmsr);
999 else
1000 update_debugctlmsr(debugctlmsr);
1001
1002 error = 0;
1003 out:
1004 local_irq_restore(irq);
1005 return error;
1006}
1007
1008static unsigned long ds_bts_control(struct bts_tracer *tracer)
1009{
1010 unsigned long control;
1011
1012 control = ds_cfg.ctl[dsf_bts];
1013 if (!(tracer->trace.ds.flags & BTS_KERNEL))
1014 control |= ds_cfg.ctl[dsf_bts_kernel];
1015 if (!(tracer->trace.ds.flags & BTS_USER))
1016 control |= ds_cfg.ctl[dsf_bts_user];
1017
1018 return control;
1019}
1020
1021void ds_resume_bts(struct bts_tracer *tracer)
1022{
1023 struct task_struct *task;
1024 unsigned long debugctlmsr;
1025 int cpu;
1026
1027 if (!tracer)
1028 return;
1029
1030 tracer->flags = tracer->trace.ds.flags;
1031
1032 task = tracer->ds.context->task;
1033 cpu = tracer->ds.context->cpu;
1034
1035 WARN_ON(!task && irqs_disabled());
1036
1037 debugctlmsr = (task ?
1038 task->thread.debugctlmsr :
1039 get_debugctlmsr_on_cpu(cpu));
1040 debugctlmsr |= ds_bts_control(tracer);
1041
1042 if (task)
1043 update_task_debugctlmsr(task, debugctlmsr);
1044 else
1045 update_debugctlmsr_on_cpu(cpu, debugctlmsr);
1046}
1047
1048int ds_resume_bts_noirq(struct bts_tracer *tracer)
1049{
1050 struct task_struct *task;
1051 unsigned long debugctlmsr, irq;
1052 int cpu, error = 0;
1053
1054 if (!tracer)
1055 return 0;
1056
1057 tracer->flags = tracer->trace.ds.flags;
1058
1059 task = tracer->ds.context->task;
1060 cpu = tracer->ds.context->cpu;
1061
1062 local_irq_save(irq);
1063
1064 error = -EPERM;
1065 if (!task && (cpu != smp_processor_id()))
1066 goto out;
1067
1068 debugctlmsr = (task ?
1069 task->thread.debugctlmsr :
1070 get_debugctlmsr());
1071 debugctlmsr |= ds_bts_control(tracer);
1072
1073 if (task)
1074 update_task_debugctlmsr(task, debugctlmsr);
1075 else
1076 update_debugctlmsr(debugctlmsr);
1077
1078 error = 0;
1079 out:
1080 local_irq_restore(irq);
1081 return error;
1082}
1083
1084static void ds_free_pebs(struct pebs_tracer *tracer)
1085{
1086 struct task_struct *task;
1087
1088 task = tracer->ds.context->task;
1089
1090 WARN_ON_ONCE(tracer->ds.context->pebs_master != tracer);
1091 tracer->ds.context->pebs_master = NULL;
1092
1093 ds_put_context(tracer->ds.context);
1094 put_tracer(task);
1095
1096 kfree(tracer);
1097}
1098
1099void ds_release_pebs(struct pebs_tracer *tracer)
1100{
1101 might_sleep();
1102
1103 if (!tracer)
1104 return;
1105
1106 ds_suspend_pebs(tracer);
1107 ds_free_pebs(tracer);
1108}
1109
1110int ds_release_pebs_noirq(struct pebs_tracer *tracer)
1111{
1112 struct task_struct *task;
1113 unsigned long irq;
1114 int error;
1115
1116 if (!tracer)
1117 return 0;
1118
1119 task = tracer->ds.context->task;
1120
1121 local_irq_save(irq);
1122
1123 error = -EPERM;
1124 if (!task &&
1125 (tracer->ds.context->cpu != smp_processor_id()))
1126 goto out;
1127
1128 error = -EPERM;
1129 if (task && (task != current))
1130 goto out;
1131
1132 ds_suspend_pebs_noirq(tracer);
1133 ds_free_pebs(tracer);
1134
1135 error = 0;
1136 out:
1137 local_irq_restore(irq);
1138 return error;
1139}
1140
1141void ds_suspend_pebs(struct pebs_tracer *tracer)
1142{
1143
1144}
1145
1146int ds_suspend_pebs_noirq(struct pebs_tracer *tracer)
1147{
1148 return 0;
1149}
1150
1151void ds_resume_pebs(struct pebs_tracer *tracer)
1152{
1153
1154}
1155
1156int ds_resume_pebs_noirq(struct pebs_tracer *tracer)
1157{
1158 return 0;
1159}
1160
1161const struct bts_trace *ds_read_bts(struct bts_tracer *tracer)
1162{
1163 if (!tracer)
1164 return NULL;
1165
1166 ds_read_config(tracer->ds.context, &tracer->trace.ds, ds_bts);
1167 return &tracer->trace;
1168}
1169
1170const struct pebs_trace *ds_read_pebs(struct pebs_tracer *tracer)
1171{
1172 if (!tracer)
1173 return NULL;
1174
1175 ds_read_config(tracer->ds.context, &tracer->trace.ds, ds_pebs);
1176
1177 tracer->trace.counters = ds_cfg.nr_counter_reset;
1178 memcpy(tracer->trace.counter_reset,
1179 tracer->ds.context->ds +
1180 (NUM_DS_PTR_FIELDS * ds_cfg.sizeof_ptr_field),
1181 ds_cfg.nr_counter_reset * PEBS_RESET_FIELD_SIZE);
1182
1183 return &tracer->trace;
1184}
1185
1186int ds_reset_bts(struct bts_tracer *tracer)
1187{
1188 if (!tracer)
1189 return -EINVAL;
1190
1191 tracer->trace.ds.top = tracer->trace.ds.begin;
1192
1193 ds_set(tracer->ds.context->ds, ds_bts, ds_index,
1194 (unsigned long)tracer->trace.ds.top);
1195
1196 return 0;
1197}
1198
1199int ds_reset_pebs(struct pebs_tracer *tracer)
1200{
1201 if (!tracer)
1202 return -EINVAL;
1203
1204 tracer->trace.ds.top = tracer->trace.ds.begin;
1205
1206 ds_set(tracer->ds.context->ds, ds_pebs, ds_index,
1207 (unsigned long)tracer->trace.ds.top);
1208
1209 return 0;
1210}
1211
1212int ds_set_pebs_reset(struct pebs_tracer *tracer,
1213 unsigned int counter, u64 value)
1214{
1215 if (!tracer)
1216 return -EINVAL;
1217
1218 if (ds_cfg.nr_counter_reset < counter)
1219 return -EINVAL;
1220
1221 *(u64 *)(tracer->ds.context->ds +
1222 (NUM_DS_PTR_FIELDS * ds_cfg.sizeof_ptr_field) +
1223 (counter * PEBS_RESET_FIELD_SIZE)) = value;
1224
1225 return 0;
1226}
1227
1228static const struct ds_configuration ds_cfg_netburst = {
1229 .name = "Netburst",
1230 .ctl[dsf_bts] = (1 << 2) | (1 << 3),
1231 .ctl[dsf_bts_kernel] = (1 << 5),
1232 .ctl[dsf_bts_user] = (1 << 6),
1233 .nr_counter_reset = 1,
1234};
1235static const struct ds_configuration ds_cfg_pentium_m = {
1236 .name = "Pentium M",
1237 .ctl[dsf_bts] = (1 << 6) | (1 << 7),
1238 .nr_counter_reset = 1,
1239};
1240static const struct ds_configuration ds_cfg_core2_atom = {
1241 .name = "Core 2/Atom",
1242 .ctl[dsf_bts] = (1 << 6) | (1 << 7),
1243 .ctl[dsf_bts_kernel] = (1 << 9),
1244 .ctl[dsf_bts_user] = (1 << 10),
1245 .nr_counter_reset = 1,
1246};
1247static const struct ds_configuration ds_cfg_core_i7 = {
1248 .name = "Core i7",
1249 .ctl[dsf_bts] = (1 << 6) | (1 << 7),
1250 .ctl[dsf_bts_kernel] = (1 << 9),
1251 .ctl[dsf_bts_user] = (1 << 10),
1252 .nr_counter_reset = 4,
1253};
1254
1255static void
1256ds_configure(const struct ds_configuration *cfg,
1257 struct cpuinfo_x86 *cpu)
1258{
1259 unsigned long nr_pebs_fields = 0;
1260
1261 printk(KERN_INFO "[ds] using %s configuration\n", cfg->name);
1262
1263#ifdef __i386__
1264 nr_pebs_fields = 10;
1265#else
1266 nr_pebs_fields = 18;
1267#endif
1268
1269 /*
1270 * Starting with version 2, architectural performance
1271 * monitoring supports a format specifier.
1272 */
1273 if ((cpuid_eax(0xa) & 0xff) > 1) {
1274 unsigned long perf_capabilities, format;
1275
1276 rdmsrl(MSR_IA32_PERF_CAPABILITIES, perf_capabilities);
1277
1278 format = (perf_capabilities >> 8) & 0xf;
1279
1280 switch (format) {
1281 case 0:
1282 nr_pebs_fields = 18;
1283 break;
1284 case 1:
1285 nr_pebs_fields = 22;
1286 break;
1287 default:
1288 printk(KERN_INFO
1289 "[ds] unknown PEBS format: %lu\n", format);
1290 nr_pebs_fields = 0;
1291 break;
1292 }
1293 }
1294
1295 memset(&ds_cfg, 0, sizeof(ds_cfg));
1296 ds_cfg = *cfg;
1297
1298 ds_cfg.sizeof_ptr_field =
1299 (cpu_has(cpu, X86_FEATURE_DTES64) ? 8 : 4);
1300
1301 ds_cfg.sizeof_rec[ds_bts] = ds_cfg.sizeof_ptr_field * 3;
1302 ds_cfg.sizeof_rec[ds_pebs] = ds_cfg.sizeof_ptr_field * nr_pebs_fields;
1303
1304 if (!cpu_has(cpu, X86_FEATURE_BTS)) {
1305 ds_cfg.sizeof_rec[ds_bts] = 0;
1306 printk(KERN_INFO "[ds] bts not available\n");
1307 }
1308 if (!cpu_has(cpu, X86_FEATURE_PEBS)) {
1309 ds_cfg.sizeof_rec[ds_pebs] = 0;
1310 printk(KERN_INFO "[ds] pebs not available\n");
1311 }
1312
1313 printk(KERN_INFO "[ds] sizes: address: %u bit, ",
1314 8 * ds_cfg.sizeof_ptr_field);
1315 printk("bts/pebs record: %u/%u bytes\n",
1316 ds_cfg.sizeof_rec[ds_bts], ds_cfg.sizeof_rec[ds_pebs]);
1317
1318 WARN_ON_ONCE(MAX_PEBS_COUNTERS < ds_cfg.nr_counter_reset);
1319}
1320
1321void __cpuinit ds_init_intel(struct cpuinfo_x86 *c)
1322{
1323 /* Only configure the first cpu. Others are identical. */
1324 if (ds_cfg.name)
1325 return;
1326
1327 switch (c->x86) {
1328 case 0x6:
1329 switch (c->x86_model) {
1330 case 0x9:
1331 case 0xd: /* Pentium M */
1332 ds_configure(&ds_cfg_pentium_m, c);
1333 break;
1334 case 0xf:
1335 case 0x17: /* Core2 */
1336 case 0x1c: /* Atom */
1337 ds_configure(&ds_cfg_core2_atom, c);
1338 break;
1339 case 0x1a: /* Core i7 */
1340 ds_configure(&ds_cfg_core_i7, c);
1341 break;
1342 default:
1343 /* Sorry, don't know about them. */
1344 break;
1345 }
1346 break;
1347 case 0xf:
1348 switch (c->x86_model) {
1349 case 0x0:
1350 case 0x1:
1351 case 0x2: /* Netburst */
1352 ds_configure(&ds_cfg_netburst, c);
1353 break;
1354 default:
1355 /* Sorry, don't know about them. */
1356 break;
1357 }
1358 break;
1359 default:
1360 /* Sorry, don't know about them. */
1361 break;
1362 }
1363}
1364
1365static inline void ds_take_timestamp(struct ds_context *context,
1366 enum bts_qualifier qualifier,
1367 struct task_struct *task)
1368{
1369 struct bts_tracer *tracer = context->bts_master;
1370 struct bts_struct ts;
1371
1372 /* Prevent compilers from reading the tracer pointer twice. */
1373 barrier();
1374
1375 if (!tracer || !(tracer->flags & BTS_TIMESTAMPS))
1376 return;
1377
1378 memset(&ts, 0, sizeof(ts));
1379 ts.qualifier = qualifier;
1380 ts.variant.event.clock = trace_clock_global();
1381 ts.variant.event.pid = task->pid;
1382
1383 bts_write(tracer, &ts);
1384}
1385
1386/*
1387 * Change the DS configuration from tracing prev to tracing next.
1388 */
1389void ds_switch_to(struct task_struct *prev, struct task_struct *next)
1390{
1391 struct ds_context *prev_ctx = prev->thread.ds_ctx;
1392 struct ds_context *next_ctx = next->thread.ds_ctx;
1393 unsigned long debugctlmsr = next->thread.debugctlmsr;
1394
1395 /* Make sure all data is read before we start. */
1396 barrier();
1397
1398 if (prev_ctx) {
1399 update_debugctlmsr(0);
1400
1401 ds_take_timestamp(prev_ctx, bts_task_departs, prev);
1402 }
1403
1404 if (next_ctx) {
1405 ds_take_timestamp(next_ctx, bts_task_arrives, next);
1406
1407 wrmsrl(MSR_IA32_DS_AREA, (unsigned long)next_ctx->ds);
1408 }
1409
1410 update_debugctlmsr(debugctlmsr);
1411}
1412
1413static __init int ds_selftest(void)
1414{
1415 if (ds_cfg.sizeof_rec[ds_bts]) {
1416 int error;
1417
1418 error = ds_selftest_bts();
1419 if (error) {
1420 WARN(1, "[ds] selftest failed. disabling bts.\n");
1421 ds_cfg.sizeof_rec[ds_bts] = 0;
1422 }
1423 }
1424
1425 if (ds_cfg.sizeof_rec[ds_pebs]) {
1426 int error;
1427
1428 error = ds_selftest_pebs();
1429 if (error) {
1430 WARN(1, "[ds] selftest failed. disabling pebs.\n");
1431 ds_cfg.sizeof_rec[ds_pebs] = 0;
1432 }
1433 }
1434
1435 return 0;
1436}
1437device_initcall(ds_selftest);
diff --git a/arch/x86/kernel/ds_selftest.c b/arch/x86/kernel/ds_selftest.c
deleted file mode 100644
index 6bc7c199ab99..000000000000
--- a/arch/x86/kernel/ds_selftest.c
+++ /dev/null
@@ -1,408 +0,0 @@
1/*
2 * Debug Store support - selftest
3 *
4 *
5 * Copyright (C) 2009 Intel Corporation.
6 * Markus Metzger <markus.t.metzger@intel.com>, 2009
7 */
8
9#include "ds_selftest.h"
10
11#include <linux/kernel.h>
12#include <linux/string.h>
13#include <linux/smp.h>
14#include <linux/cpu.h>
15
16#include <asm/ds.h>
17
18
19#define BUFFER_SIZE 521 /* Intentionally chose an odd size. */
20#define SMALL_BUFFER_SIZE 24 /* A single bts entry. */
21
22struct ds_selftest_bts_conf {
23 struct bts_tracer *tracer;
24 int error;
25 int (*suspend)(struct bts_tracer *);
26 int (*resume)(struct bts_tracer *);
27};
28
29static int ds_selftest_bts_consistency(const struct bts_trace *trace)
30{
31 int error = 0;
32
33 if (!trace) {
34 printk(KERN_CONT "failed to access trace...");
35 /* Bail out. Other tests are pointless. */
36 return -1;
37 }
38
39 if (!trace->read) {
40 printk(KERN_CONT "bts read not available...");
41 error = -1;
42 }
43
44 /* Do some sanity checks on the trace configuration. */
45 if (!trace->ds.n) {
46 printk(KERN_CONT "empty bts buffer...");
47 error = -1;
48 }
49 if (!trace->ds.size) {
50 printk(KERN_CONT "bad bts trace setup...");
51 error = -1;
52 }
53 if (trace->ds.end !=
54 (char *)trace->ds.begin + (trace->ds.n * trace->ds.size)) {
55 printk(KERN_CONT "bad bts buffer setup...");
56 error = -1;
57 }
58 /*
59 * We allow top in [begin; end], since its not clear when the
60 * overflow adjustment happens: after the increment or before the
61 * write.
62 */
63 if ((trace->ds.top < trace->ds.begin) ||
64 (trace->ds.end < trace->ds.top)) {
65 printk(KERN_CONT "bts top out of bounds...");
66 error = -1;
67 }
68
69 return error;
70}
71
72static int ds_selftest_bts_read(struct bts_tracer *tracer,
73 const struct bts_trace *trace,
74 const void *from, const void *to)
75{
76 const unsigned char *at;
77
78 /*
79 * Check a few things which do not belong to this test.
80 * They should be covered by other tests.
81 */
82 if (!trace)
83 return -1;
84
85 if (!trace->read)
86 return -1;
87
88 if (to < from)
89 return -1;
90
91 if (from < trace->ds.begin)
92 return -1;
93
94 if (trace->ds.end < to)
95 return -1;
96
97 if (!trace->ds.size)
98 return -1;
99
100 /* Now to the test itself. */
101 for (at = from; (void *)at < to; at += trace->ds.size) {
102 struct bts_struct bts;
103 unsigned long index;
104 int error;
105
106 if (((void *)at - trace->ds.begin) % trace->ds.size) {
107 printk(KERN_CONT
108 "read from non-integer index...");
109 return -1;
110 }
111 index = ((void *)at - trace->ds.begin) / trace->ds.size;
112
113 memset(&bts, 0, sizeof(bts));
114 error = trace->read(tracer, at, &bts);
115 if (error < 0) {
116 printk(KERN_CONT
117 "error reading bts trace at [%lu] (0x%p)...",
118 index, at);
119 return error;
120 }
121
122 switch (bts.qualifier) {
123 case BTS_BRANCH:
124 break;
125 default:
126 printk(KERN_CONT
127 "unexpected bts entry %llu at [%lu] (0x%p)...",
128 bts.qualifier, index, at);
129 return -1;
130 }
131 }
132
133 return 0;
134}
135
136static void ds_selftest_bts_cpu(void *arg)
137{
138 struct ds_selftest_bts_conf *conf = arg;
139 const struct bts_trace *trace;
140 void *top;
141
142 if (IS_ERR(conf->tracer)) {
143 conf->error = PTR_ERR(conf->tracer);
144 conf->tracer = NULL;
145
146 printk(KERN_CONT
147 "initialization failed (err: %d)...", conf->error);
148 return;
149 }
150
151 /* We should meanwhile have enough trace. */
152 conf->error = conf->suspend(conf->tracer);
153 if (conf->error < 0)
154 return;
155
156 /* Let's see if we can access the trace. */
157 trace = ds_read_bts(conf->tracer);
158
159 conf->error = ds_selftest_bts_consistency(trace);
160 if (conf->error < 0)
161 return;
162
163 /* If everything went well, we should have a few trace entries. */
164 if (trace->ds.top == trace->ds.begin) {
165 /*
166 * It is possible but highly unlikely that we got a
167 * buffer overflow and end up at exactly the same
168 * position we started from.
169 * Let's issue a warning, but continue.
170 */
171 printk(KERN_CONT "no trace/overflow...");
172 }
173
174 /* Let's try to read the trace we collected. */
175 conf->error =
176 ds_selftest_bts_read(conf->tracer, trace,
177 trace->ds.begin, trace->ds.top);
178 if (conf->error < 0)
179 return;
180
181 /*
182 * Let's read the trace again.
183 * Since we suspended tracing, we should get the same result.
184 */
185 top = trace->ds.top;
186
187 trace = ds_read_bts(conf->tracer);
188 conf->error = ds_selftest_bts_consistency(trace);
189 if (conf->error < 0)
190 return;
191
192 if (top != trace->ds.top) {
193 printk(KERN_CONT "suspend not working...");
194 conf->error = -1;
195 return;
196 }
197
198 /* Let's collect some more trace - see if resume is working. */
199 conf->error = conf->resume(conf->tracer);
200 if (conf->error < 0)
201 return;
202
203 conf->error = conf->suspend(conf->tracer);
204 if (conf->error < 0)
205 return;
206
207 trace = ds_read_bts(conf->tracer);
208
209 conf->error = ds_selftest_bts_consistency(trace);
210 if (conf->error < 0)
211 return;
212
213 if (trace->ds.top == top) {
214 /*
215 * It is possible but highly unlikely that we got a
216 * buffer overflow and end up at exactly the same
217 * position we started from.
218 * Let's issue a warning and check the full trace.
219 */
220 printk(KERN_CONT
221 "no resume progress/overflow...");
222
223 conf->error =
224 ds_selftest_bts_read(conf->tracer, trace,
225 trace->ds.begin, trace->ds.end);
226 } else if (trace->ds.top < top) {
227 /*
228 * We had a buffer overflow - the entire buffer should
229 * contain trace records.
230 */
231 conf->error =
232 ds_selftest_bts_read(conf->tracer, trace,
233 trace->ds.begin, trace->ds.end);
234 } else {
235 /*
236 * It is quite likely that the buffer did not overflow.
237 * Let's just check the delta trace.
238 */
239 conf->error =
240 ds_selftest_bts_read(conf->tracer, trace, top,
241 trace->ds.top);
242 }
243 if (conf->error < 0)
244 return;
245
246 conf->error = 0;
247}
248
249static int ds_suspend_bts_wrap(struct bts_tracer *tracer)
250{
251 ds_suspend_bts(tracer);
252 return 0;
253}
254
255static int ds_resume_bts_wrap(struct bts_tracer *tracer)
256{
257 ds_resume_bts(tracer);
258 return 0;
259}
260
261static void ds_release_bts_noirq_wrap(void *tracer)
262{
263 (void)ds_release_bts_noirq(tracer);
264}
265
266static int ds_selftest_bts_bad_release_noirq(int cpu,
267 struct bts_tracer *tracer)
268{
269 int error = -EPERM;
270
271 /* Try to release the tracer on the wrong cpu. */
272 get_cpu();
273 if (cpu != smp_processor_id()) {
274 error = ds_release_bts_noirq(tracer);
275 if (error != -EPERM)
276 printk(KERN_CONT "release on wrong cpu...");
277 }
278 put_cpu();
279
280 return error ? 0 : -1;
281}
282
283static int ds_selftest_bts_bad_request_cpu(int cpu, void *buffer)
284{
285 struct bts_tracer *tracer;
286 int error;
287
288 /* Try to request cpu tracing while task tracing is active. */
289 tracer = ds_request_bts_cpu(cpu, buffer, BUFFER_SIZE, NULL,
290 (size_t)-1, BTS_KERNEL);
291 error = PTR_ERR(tracer);
292 if (!IS_ERR(tracer)) {
293 ds_release_bts(tracer);
294 error = 0;
295 }
296
297 if (error != -EPERM)
298 printk(KERN_CONT "cpu/task tracing overlap...");
299
300 return error ? 0 : -1;
301}
302
303static int ds_selftest_bts_bad_request_task(void *buffer)
304{
305 struct bts_tracer *tracer;
306 int error;
307
308 /* Try to request cpu tracing while task tracing is active. */
309 tracer = ds_request_bts_task(current, buffer, BUFFER_SIZE, NULL,
310 (size_t)-1, BTS_KERNEL);
311 error = PTR_ERR(tracer);
312 if (!IS_ERR(tracer)) {
313 error = 0;
314 ds_release_bts(tracer);
315 }
316
317 if (error != -EPERM)
318 printk(KERN_CONT "task/cpu tracing overlap...");
319
320 return error ? 0 : -1;
321}
322
323int ds_selftest_bts(void)
324{
325 struct ds_selftest_bts_conf conf;
326 unsigned char buffer[BUFFER_SIZE], *small_buffer;
327 unsigned long irq;
328 int cpu;
329
330 printk(KERN_INFO "[ds] bts selftest...");
331 conf.error = 0;
332
333 small_buffer = (unsigned char *)ALIGN((unsigned long)buffer, 8) + 8;
334
335 get_online_cpus();
336 for_each_online_cpu(cpu) {
337 conf.suspend = ds_suspend_bts_wrap;
338 conf.resume = ds_resume_bts_wrap;
339 conf.tracer =
340 ds_request_bts_cpu(cpu, buffer, BUFFER_SIZE,
341 NULL, (size_t)-1, BTS_KERNEL);
342 ds_selftest_bts_cpu(&conf);
343 if (conf.error >= 0)
344 conf.error = ds_selftest_bts_bad_request_task(buffer);
345 ds_release_bts(conf.tracer);
346 if (conf.error < 0)
347 goto out;
348
349 conf.suspend = ds_suspend_bts_noirq;
350 conf.resume = ds_resume_bts_noirq;
351 conf.tracer =
352 ds_request_bts_cpu(cpu, buffer, BUFFER_SIZE,
353 NULL, (size_t)-1, BTS_KERNEL);
354 smp_call_function_single(cpu, ds_selftest_bts_cpu, &conf, 1);
355 if (conf.error >= 0) {
356 conf.error =
357 ds_selftest_bts_bad_release_noirq(cpu,
358 conf.tracer);
359 /* We must not release the tracer twice. */
360 if (conf.error < 0)
361 conf.tracer = NULL;
362 }
363 if (conf.error >= 0)
364 conf.error = ds_selftest_bts_bad_request_task(buffer);
365 smp_call_function_single(cpu, ds_release_bts_noirq_wrap,
366 conf.tracer, 1);
367 if (conf.error < 0)
368 goto out;
369 }
370
371 conf.suspend = ds_suspend_bts_wrap;
372 conf.resume = ds_resume_bts_wrap;
373 conf.tracer =
374 ds_request_bts_task(current, buffer, BUFFER_SIZE,
375 NULL, (size_t)-1, BTS_KERNEL);
376 ds_selftest_bts_cpu(&conf);
377 if (conf.error >= 0)
378 conf.error = ds_selftest_bts_bad_request_cpu(0, buffer);
379 ds_release_bts(conf.tracer);
380 if (conf.error < 0)
381 goto out;
382
383 conf.suspend = ds_suspend_bts_noirq;
384 conf.resume = ds_resume_bts_noirq;
385 conf.tracer =
386 ds_request_bts_task(current, small_buffer, SMALL_BUFFER_SIZE,
387 NULL, (size_t)-1, BTS_KERNEL);
388 local_irq_save(irq);
389 ds_selftest_bts_cpu(&conf);
390 if (conf.error >= 0)
391 conf.error = ds_selftest_bts_bad_request_cpu(0, buffer);
392 ds_release_bts_noirq(conf.tracer);
393 local_irq_restore(irq);
394 if (conf.error < 0)
395 goto out;
396
397 conf.error = 0;
398 out:
399 put_online_cpus();
400 printk(KERN_CONT "%s.\n", (conf.error ? "failed" : "passed"));
401
402 return conf.error;
403}
404
405int ds_selftest_pebs(void)
406{
407 return 0;
408}
diff --git a/arch/x86/kernel/ds_selftest.h b/arch/x86/kernel/ds_selftest.h
deleted file mode 100644
index 2ba8745c6663..000000000000
--- a/arch/x86/kernel/ds_selftest.h
+++ /dev/null
@@ -1,15 +0,0 @@
1/*
2 * Debug Store support - selftest
3 *
4 *
5 * Copyright (C) 2009 Intel Corporation.
6 * Markus Metzger <markus.t.metzger@intel.com>, 2009
7 */
8
9#ifdef CONFIG_X86_DS_SELFTEST
10extern int ds_selftest_bts(void);
11extern int ds_selftest_pebs(void);
12#else
13static inline int ds_selftest_bts(void) { return 0; }
14static inline int ds_selftest_pebs(void) { return 0; }
15#endif
diff --git a/arch/x86/kernel/dumpstack.c b/arch/x86/kernel/dumpstack.c
index 6d817554780a..c89a386930b7 100644
--- a/arch/x86/kernel/dumpstack.c
+++ b/arch/x86/kernel/dumpstack.c
@@ -224,11 +224,6 @@ unsigned __kprobes long oops_begin(void)
224 int cpu; 224 int cpu;
225 unsigned long flags; 225 unsigned long flags;
226 226
227 /* notify the hw-branch tracer so it may disable tracing and
228 add the last trace to the trace buffer -
229 the earlier this happens, the more useful the trace. */
230 trace_hw_branch_oops();
231
232 oops_enter(); 227 oops_enter();
233 228
234 /* racy, but better than risking deadlock. */ 229 /* racy, but better than risking deadlock. */
diff --git a/arch/x86/kernel/kprobes.c b/arch/x86/kernel/kprobes.c
index b43bbaebe2c0..7a880ad3a208 100644
--- a/arch/x86/kernel/kprobes.c
+++ b/arch/x86/kernel/kprobes.c
@@ -422,14 +422,12 @@ static void __kprobes set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
422 422
423static void __kprobes clear_btf(void) 423static void __kprobes clear_btf(void)
424{ 424{
425 if (test_thread_flag(TIF_DEBUGCTLMSR)) 425 /* XXX */
426 update_debugctlmsr(0);
427} 426}
428 427
429static void __kprobes restore_btf(void) 428static void __kprobes restore_btf(void)
430{ 429{
431 if (test_thread_flag(TIF_DEBUGCTLMSR)) 430 /* XXX */
432 update_debugctlmsr(current->thread.debugctlmsr);
433} 431}
434 432
435void __kprobes arch_prepare_kretprobe(struct kretprobe_instance *ri, 433void __kprobes arch_prepare_kretprobe(struct kretprobe_instance *ri,
diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c
index ad9540676fcc..1a60beb32ede 100644
--- a/arch/x86/kernel/process.c
+++ b/arch/x86/kernel/process.c
@@ -20,7 +20,6 @@
20#include <asm/idle.h> 20#include <asm/idle.h>
21#include <asm/uaccess.h> 21#include <asm/uaccess.h>
22#include <asm/i387.h> 22#include <asm/i387.h>
23#include <asm/ds.h>
24#include <asm/debugreg.h> 23#include <asm/debugreg.h>
25 24
26unsigned long idle_halt; 25unsigned long idle_halt;
@@ -50,8 +49,6 @@ void free_thread_xstate(struct task_struct *tsk)
50 kmem_cache_free(task_xstate_cachep, tsk->thread.xstate); 49 kmem_cache_free(task_xstate_cachep, tsk->thread.xstate);
51 tsk->thread.xstate = NULL; 50 tsk->thread.xstate = NULL;
52 } 51 }
53
54 WARN(tsk->thread.ds_ctx, "leaking DS context\n");
55} 52}
56 53
57void free_thread_info(struct thread_info *ti) 54void free_thread_info(struct thread_info *ti)
@@ -198,12 +195,6 @@ void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
198 prev = &prev_p->thread; 195 prev = &prev_p->thread;
199 next = &next_p->thread; 196 next = &next_p->thread;
200 197
201 if (test_tsk_thread_flag(next_p, TIF_DS_AREA_MSR) ||
202 test_tsk_thread_flag(prev_p, TIF_DS_AREA_MSR))
203 ds_switch_to(prev_p, next_p);
204 else if (next->debugctlmsr != prev->debugctlmsr)
205 update_debugctlmsr(next->debugctlmsr);
206
207 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^ 198 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
208 test_tsk_thread_flag(next_p, TIF_NOTSC)) { 199 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
209 /* prev and next are different */ 200 /* prev and next are different */
diff --git a/arch/x86/kernel/process_32.c b/arch/x86/kernel/process_32.c
index f6c62667e30c..75090c589b7a 100644
--- a/arch/x86/kernel/process_32.c
+++ b/arch/x86/kernel/process_32.c
@@ -55,7 +55,6 @@
55#include <asm/cpu.h> 55#include <asm/cpu.h>
56#include <asm/idle.h> 56#include <asm/idle.h>
57#include <asm/syscalls.h> 57#include <asm/syscalls.h>
58#include <asm/ds.h>
59#include <asm/debugreg.h> 58#include <asm/debugreg.h>
60 59
61asmlinkage void ret_from_fork(void) __asm__("ret_from_fork"); 60asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
@@ -238,13 +237,6 @@ int copy_thread(unsigned long clone_flags, unsigned long sp,
238 kfree(p->thread.io_bitmap_ptr); 237 kfree(p->thread.io_bitmap_ptr);
239 p->thread.io_bitmap_max = 0; 238 p->thread.io_bitmap_max = 0;
240 } 239 }
241
242 clear_tsk_thread_flag(p, TIF_DS_AREA_MSR);
243 p->thread.ds_ctx = NULL;
244
245 clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR);
246 p->thread.debugctlmsr = 0;
247
248 return err; 240 return err;
249} 241}
250 242
diff --git a/arch/x86/kernel/process_64.c b/arch/x86/kernel/process_64.c
index dc9690b4c4cc..cc4258f2beb5 100644
--- a/arch/x86/kernel/process_64.c
+++ b/arch/x86/kernel/process_64.c
@@ -49,7 +49,6 @@
49#include <asm/ia32.h> 49#include <asm/ia32.h>
50#include <asm/idle.h> 50#include <asm/idle.h>
51#include <asm/syscalls.h> 51#include <asm/syscalls.h>
52#include <asm/ds.h>
53#include <asm/debugreg.h> 52#include <asm/debugreg.h>
54 53
55asmlinkage extern void ret_from_fork(void); 54asmlinkage extern void ret_from_fork(void);
@@ -313,13 +312,6 @@ int copy_thread(unsigned long clone_flags, unsigned long sp,
313 if (err) 312 if (err)
314 goto out; 313 goto out;
315 } 314 }
316
317 clear_tsk_thread_flag(p, TIF_DS_AREA_MSR);
318 p->thread.ds_ctx = NULL;
319
320 clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR);
321 p->thread.debugctlmsr = 0;
322
323 err = 0; 315 err = 0;
324out: 316out:
325 if (err && p->thread.io_bitmap_ptr) { 317 if (err && p->thread.io_bitmap_ptr) {
diff --git a/arch/x86/kernel/ptrace.c b/arch/x86/kernel/ptrace.c
index a503b1fd04e5..f2fd3b80e565 100644
--- a/arch/x86/kernel/ptrace.c
+++ b/arch/x86/kernel/ptrace.c
@@ -2,9 +2,6 @@
2/* 2/*
3 * Pentium III FXSR, SSE support 3 * Pentium III FXSR, SSE support
4 * Gareth Hughes <gareth@valinux.com>, May 2000 4 * Gareth Hughes <gareth@valinux.com>, May 2000
5 *
6 * BTS tracing
7 * Markus Metzger <markus.t.metzger@intel.com>, Dec 2007
8 */ 5 */
9 6
10#include <linux/kernel.h> 7#include <linux/kernel.h>
@@ -21,7 +18,6 @@
21#include <linux/audit.h> 18#include <linux/audit.h>
22#include <linux/seccomp.h> 19#include <linux/seccomp.h>
23#include <linux/signal.h> 20#include <linux/signal.h>
24#include <linux/workqueue.h>
25#include <linux/perf_event.h> 21#include <linux/perf_event.h>
26#include <linux/hw_breakpoint.h> 22#include <linux/hw_breakpoint.h>
27 23
@@ -35,7 +31,6 @@
35#include <asm/desc.h> 31#include <asm/desc.h>
36#include <asm/prctl.h> 32#include <asm/prctl.h>
37#include <asm/proto.h> 33#include <asm/proto.h>
38#include <asm/ds.h>
39#include <asm/hw_breakpoint.h> 34#include <asm/hw_breakpoint.h>
40 35
41#include "tls.h" 36#include "tls.h"
@@ -788,342 +783,6 @@ static int ioperm_get(struct task_struct *target,
788 0, IO_BITMAP_BYTES); 783 0, IO_BITMAP_BYTES);
789} 784}
790 785
791#ifdef CONFIG_X86_PTRACE_BTS
792/*
793 * A branch trace store context.
794 *
795 * Contexts may only be installed by ptrace_bts_config() and only for
796 * ptraced tasks.
797 *
798 * Contexts are destroyed when the tracee is detached from the tracer.
799 * The actual destruction work requires interrupts enabled, so the
800 * work is deferred and will be scheduled during __ptrace_unlink().
801 *
802 * Contexts hold an additional task_struct reference on the traced
803 * task, as well as a reference on the tracer's mm.
804 *
805 * Ptrace already holds a task_struct for the duration of ptrace operations,
806 * but since destruction is deferred, it may be executed after both
807 * tracer and tracee exited.
808 */
809struct bts_context {
810 /* The branch trace handle. */
811 struct bts_tracer *tracer;
812
813 /* The buffer used to store the branch trace and its size. */
814 void *buffer;
815 unsigned int size;
816
817 /* The mm that paid for the above buffer. */
818 struct mm_struct *mm;
819
820 /* The task this context belongs to. */
821 struct task_struct *task;
822
823 /* The signal to send on a bts buffer overflow. */
824 unsigned int bts_ovfl_signal;
825
826 /* The work struct to destroy a context. */
827 struct work_struct work;
828};
829
830static int alloc_bts_buffer(struct bts_context *context, unsigned int size)
831{
832 void *buffer = NULL;
833 int err = -ENOMEM;
834
835 err = account_locked_memory(current->mm, current->signal->rlim, size);
836 if (err < 0)
837 return err;
838
839 buffer = kzalloc(size, GFP_KERNEL);
840 if (!buffer)
841 goto out_refund;
842
843 context->buffer = buffer;
844 context->size = size;
845 context->mm = get_task_mm(current);
846
847 return 0;
848
849 out_refund:
850 refund_locked_memory(current->mm, size);
851 return err;
852}
853
854static inline void free_bts_buffer(struct bts_context *context)
855{
856 if (!context->buffer)
857 return;
858
859 kfree(context->buffer);
860 context->buffer = NULL;
861
862 refund_locked_memory(context->mm, context->size);
863 context->size = 0;
864
865 mmput(context->mm);
866 context->mm = NULL;
867}
868
869static void free_bts_context_work(struct work_struct *w)
870{
871 struct bts_context *context;
872
873 context = container_of(w, struct bts_context, work);
874
875 ds_release_bts(context->tracer);
876 put_task_struct(context->task);
877 free_bts_buffer(context);
878 kfree(context);
879}
880
881static inline void free_bts_context(struct bts_context *context)
882{
883 INIT_WORK(&context->work, free_bts_context_work);
884 schedule_work(&context->work);
885}
886
887static inline struct bts_context *alloc_bts_context(struct task_struct *task)
888{
889 struct bts_context *context = kzalloc(sizeof(*context), GFP_KERNEL);
890 if (context) {
891 context->task = task;
892 task->bts = context;
893
894 get_task_struct(task);
895 }
896
897 return context;
898}
899
900static int ptrace_bts_read_record(struct task_struct *child, size_t index,
901 struct bts_struct __user *out)
902{
903 struct bts_context *context;
904 const struct bts_trace *trace;
905 struct bts_struct bts;
906 const unsigned char *at;
907 int error;
908
909 context = child->bts;
910 if (!context)
911 return -ESRCH;
912
913 trace = ds_read_bts(context->tracer);
914 if (!trace)
915 return -ESRCH;
916
917 at = trace->ds.top - ((index + 1) * trace->ds.size);
918 if ((void *)at < trace->ds.begin)
919 at += (trace->ds.n * trace->ds.size);
920
921 if (!trace->read)
922 return -EOPNOTSUPP;
923
924 error = trace->read(context->tracer, at, &bts);
925 if (error < 0)
926 return error;
927
928 if (copy_to_user(out, &bts, sizeof(bts)))
929 return -EFAULT;
930
931 return sizeof(bts);
932}
933
934static int ptrace_bts_drain(struct task_struct *child,
935 long size,
936 struct bts_struct __user *out)
937{
938 struct bts_context *context;
939 const struct bts_trace *trace;
940 const unsigned char *at;
941 int error, drained = 0;
942
943 context = child->bts;
944 if (!context)
945 return -ESRCH;
946
947 trace = ds_read_bts(context->tracer);
948 if (!trace)
949 return -ESRCH;
950
951 if (!trace->read)
952 return -EOPNOTSUPP;
953
954 if (size < (trace->ds.top - trace->ds.begin))
955 return -EIO;
956
957 for (at = trace->ds.begin; (void *)at < trace->ds.top;
958 out++, drained++, at += trace->ds.size) {
959 struct bts_struct bts;
960
961 error = trace->read(context->tracer, at, &bts);
962 if (error < 0)
963 return error;
964
965 if (copy_to_user(out, &bts, sizeof(bts)))
966 return -EFAULT;
967 }
968
969 memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
970
971 error = ds_reset_bts(context->tracer);
972 if (error < 0)
973 return error;
974
975 return drained;
976}
977
978static int ptrace_bts_config(struct task_struct *child,
979 long cfg_size,
980 const struct ptrace_bts_config __user *ucfg)
981{
982 struct bts_context *context;
983 struct ptrace_bts_config cfg;
984 unsigned int flags = 0;
985
986 if (cfg_size < sizeof(cfg))
987 return -EIO;
988
989 if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
990 return -EFAULT;
991
992 context = child->bts;
993 if (!context)
994 context = alloc_bts_context(child);
995 if (!context)
996 return -ENOMEM;
997
998 if (cfg.flags & PTRACE_BTS_O_SIGNAL) {
999 if (!cfg.signal)
1000 return -EINVAL;
1001
1002 return -EOPNOTSUPP;
1003 context->bts_ovfl_signal = cfg.signal;
1004 }
1005
1006 ds_release_bts(context->tracer);
1007 context->tracer = NULL;
1008
1009 if ((cfg.flags & PTRACE_BTS_O_ALLOC) && (cfg.size != context->size)) {
1010 int err;
1011
1012 free_bts_buffer(context);
1013 if (!cfg.size)
1014 return 0;
1015
1016 err = alloc_bts_buffer(context, cfg.size);
1017 if (err < 0)
1018 return err;
1019 }
1020
1021 if (cfg.flags & PTRACE_BTS_O_TRACE)
1022 flags |= BTS_USER;
1023
1024 if (cfg.flags & PTRACE_BTS_O_SCHED)
1025 flags |= BTS_TIMESTAMPS;
1026
1027 context->tracer =
1028 ds_request_bts_task(child, context->buffer, context->size,
1029 NULL, (size_t)-1, flags);
1030 if (unlikely(IS_ERR(context->tracer))) {
1031 int error = PTR_ERR(context->tracer);
1032
1033 free_bts_buffer(context);
1034 context->tracer = NULL;
1035 return error;
1036 }
1037
1038 return sizeof(cfg);
1039}
1040
1041static int ptrace_bts_status(struct task_struct *child,
1042 long cfg_size,
1043 struct ptrace_bts_config __user *ucfg)
1044{
1045 struct bts_context *context;
1046 const struct bts_trace *trace;
1047 struct ptrace_bts_config cfg;
1048
1049 context = child->bts;
1050 if (!context)
1051 return -ESRCH;
1052
1053 if (cfg_size < sizeof(cfg))
1054 return -EIO;
1055
1056 trace = ds_read_bts(context->tracer);
1057 if (!trace)
1058 return -ESRCH;
1059
1060 memset(&cfg, 0, sizeof(cfg));
1061 cfg.size = trace->ds.end - trace->ds.begin;
1062 cfg.signal = context->bts_ovfl_signal;
1063 cfg.bts_size = sizeof(struct bts_struct);
1064
1065 if (cfg.signal)
1066 cfg.flags |= PTRACE_BTS_O_SIGNAL;
1067
1068 if (trace->ds.flags & BTS_USER)
1069 cfg.flags |= PTRACE_BTS_O_TRACE;
1070
1071 if (trace->ds.flags & BTS_TIMESTAMPS)
1072 cfg.flags |= PTRACE_BTS_O_SCHED;
1073
1074 if (copy_to_user(ucfg, &cfg, sizeof(cfg)))
1075 return -EFAULT;
1076
1077 return sizeof(cfg);
1078}
1079
1080static int ptrace_bts_clear(struct task_struct *child)
1081{
1082 struct bts_context *context;
1083 const struct bts_trace *trace;
1084
1085 context = child->bts;
1086 if (!context)
1087 return -ESRCH;
1088
1089 trace = ds_read_bts(context->tracer);
1090 if (!trace)
1091 return -ESRCH;
1092
1093 memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
1094
1095 return ds_reset_bts(context->tracer);
1096}
1097
1098static int ptrace_bts_size(struct task_struct *child)
1099{
1100 struct bts_context *context;
1101 const struct bts_trace *trace;
1102
1103 context = child->bts;
1104 if (!context)
1105 return -ESRCH;
1106
1107 trace = ds_read_bts(context->tracer);
1108 if (!trace)
1109 return -ESRCH;
1110
1111 return (trace->ds.top - trace->ds.begin) / trace->ds.size;
1112}
1113
1114/*
1115 * Called from __ptrace_unlink() after the child has been moved back
1116 * to its original parent.
1117 */
1118void ptrace_bts_untrace(struct task_struct *child)
1119{
1120 if (unlikely(child->bts)) {
1121 free_bts_context(child->bts);
1122 child->bts = NULL;
1123 }
1124}
1125#endif /* CONFIG_X86_PTRACE_BTS */
1126
1127/* 786/*
1128 * Called by kernel/ptrace.c when detaching.. 787 * Called by kernel/ptrace.c when detaching..
1129 * 788 *
@@ -1251,39 +910,6 @@ long arch_ptrace(struct task_struct *child, long request, long addr, long data)
1251 break; 910 break;
1252#endif 911#endif
1253 912
1254 /*
1255 * These bits need more cooking - not enabled yet:
1256 */
1257#ifdef CONFIG_X86_PTRACE_BTS
1258 case PTRACE_BTS_CONFIG:
1259 ret = ptrace_bts_config
1260 (child, data, (struct ptrace_bts_config __user *)addr);
1261 break;
1262
1263 case PTRACE_BTS_STATUS:
1264 ret = ptrace_bts_status
1265 (child, data, (struct ptrace_bts_config __user *)addr);
1266 break;
1267
1268 case PTRACE_BTS_SIZE:
1269 ret = ptrace_bts_size(child);
1270 break;
1271
1272 case PTRACE_BTS_GET:
1273 ret = ptrace_bts_read_record
1274 (child, data, (struct bts_struct __user *) addr);
1275 break;
1276
1277 case PTRACE_BTS_CLEAR:
1278 ret = ptrace_bts_clear(child);
1279 break;
1280
1281 case PTRACE_BTS_DRAIN:
1282 ret = ptrace_bts_drain
1283 (child, data, (struct bts_struct __user *) addr);
1284 break;
1285#endif /* CONFIG_X86_PTRACE_BTS */
1286
1287 default: 913 default:
1288 ret = ptrace_request(child, request, addr, data); 914 ret = ptrace_request(child, request, addr, data);
1289 break; 915 break;
@@ -1543,14 +1169,6 @@ long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1543 1169
1544 case PTRACE_GET_THREAD_AREA: 1170 case PTRACE_GET_THREAD_AREA:
1545 case PTRACE_SET_THREAD_AREA: 1171 case PTRACE_SET_THREAD_AREA:
1546#ifdef CONFIG_X86_PTRACE_BTS
1547 case PTRACE_BTS_CONFIG:
1548 case PTRACE_BTS_STATUS:
1549 case PTRACE_BTS_SIZE:
1550 case PTRACE_BTS_GET:
1551 case PTRACE_BTS_CLEAR:
1552 case PTRACE_BTS_DRAIN:
1553#endif /* CONFIG_X86_PTRACE_BTS */
1554 return arch_ptrace(child, request, addr, data); 1172 return arch_ptrace(child, request, addr, data);
1555 1173
1556 default: 1174 default:
diff --git a/arch/x86/kernel/step.c b/arch/x86/kernel/step.c
index 3149032ff107..7beba0769a8c 100644
--- a/arch/x86/kernel/step.c
+++ b/arch/x86/kernel/step.c
@@ -158,22 +158,6 @@ static int enable_single_step(struct task_struct *child)
158} 158}
159 159
160/* 160/*
161 * Install this value in MSR_IA32_DEBUGCTLMSR whenever child is running.
162 */
163static void write_debugctlmsr(struct task_struct *child, unsigned long val)
164{
165 if (child->thread.debugctlmsr == val)
166 return;
167
168 child->thread.debugctlmsr = val;
169
170 if (child != current)
171 return;
172
173 update_debugctlmsr(val);
174}
175
176/*
177 * Enable single or block step. 161 * Enable single or block step.
178 */ 162 */
179static void enable_step(struct task_struct *child, bool block) 163static void enable_step(struct task_struct *child, bool block)
@@ -185,17 +169,9 @@ static void enable_step(struct task_struct *child, bool block)
185 * So noone should try to use debugger block stepping in a program 169 * So noone should try to use debugger block stepping in a program
186 * that uses user-mode single stepping itself. 170 * that uses user-mode single stepping itself.
187 */ 171 */
188 if (enable_single_step(child) && block) { 172 if (!enable_single_step(child))
189 set_tsk_thread_flag(child, TIF_DEBUGCTLMSR); 173 return;
190 write_debugctlmsr(child, 174 /* XXX */
191 child->thread.debugctlmsr | DEBUGCTLMSR_BTF);
192 } else {
193 write_debugctlmsr(child,
194 child->thread.debugctlmsr & ~DEBUGCTLMSR_BTF);
195
196 if (!child->thread.debugctlmsr)
197 clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
198 }
199} 175}
200 176
201void user_enable_single_step(struct task_struct *child) 177void user_enable_single_step(struct task_struct *child)
@@ -213,11 +189,7 @@ void user_disable_single_step(struct task_struct *child)
213 /* 189 /*
214 * Make sure block stepping (BTF) is disabled. 190 * Make sure block stepping (BTF) is disabled.
215 */ 191 */
216 write_debugctlmsr(child, 192 /* XXX */
217 child->thread.debugctlmsr & ~DEBUGCTLMSR_BTF);
218
219 if (!child->thread.debugctlmsr)
220 clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
221 193
222 /* Always clear TIF_SINGLESTEP... */ 194 /* Always clear TIF_SINGLESTEP... */
223 clear_tsk_thread_flag(child, TIF_SINGLESTEP); 195 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
diff --git a/arch/x86/kernel/traps.c b/arch/x86/kernel/traps.c
index 1168e4454188..e3da5d726a37 100644
--- a/arch/x86/kernel/traps.c
+++ b/arch/x86/kernel/traps.c
@@ -543,11 +543,6 @@ dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
543 543
544 /* DR6 may or may not be cleared by the CPU */ 544 /* DR6 may or may not be cleared by the CPU */
545 set_debugreg(0, 6); 545 set_debugreg(0, 6);
546 /*
547 * The processor cleared BTF, so don't mark that we need it set.
548 */
549 clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
550 tsk->thread.debugctlmsr = 0;
551 546
552 /* Store the virtualized DR6 value */ 547 /* Store the virtualized DR6 value */
553 tsk->thread.debugreg6 = dr6; 548 tsk->thread.debugreg6 = dr6;
diff --git a/include/linux/ftrace.h b/include/linux/ftrace.h
index 01e6adea07ec..cc12b3c556b3 100644
--- a/include/linux/ftrace.h
+++ b/include/linux/ftrace.h
@@ -504,18 +504,6 @@ extern int ftrace_dump_on_oops;
504#define INIT_TRACE_RECURSION 504#define INIT_TRACE_RECURSION
505#endif 505#endif
506 506
507#ifdef CONFIG_HW_BRANCH_TRACER
508
509void trace_hw_branch(u64 from, u64 to);
510void trace_hw_branch_oops(void);
511
512#else /* CONFIG_HW_BRANCH_TRACER */
513
514static inline void trace_hw_branch(u64 from, u64 to) {}
515static inline void trace_hw_branch_oops(void) {}
516
517#endif /* CONFIG_HW_BRANCH_TRACER */
518
519#ifdef CONFIG_FTRACE_SYSCALLS 507#ifdef CONFIG_FTRACE_SYSCALLS
520 508
521unsigned long arch_syscall_addr(int nr); 509unsigned long arch_syscall_addr(int nr);
diff --git a/include/linux/mm.h b/include/linux/mm.h
index e70f21beb4b4..c8442b655111 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -19,7 +19,6 @@ struct anon_vma;
19struct file_ra_state; 19struct file_ra_state;
20struct user_struct; 20struct user_struct;
21struct writeback_control; 21struct writeback_control;
22struct rlimit;
23 22
24#ifndef CONFIG_DISCONTIGMEM /* Don't use mapnrs, do it properly */ 23#ifndef CONFIG_DISCONTIGMEM /* Don't use mapnrs, do it properly */
25extern unsigned long max_mapnr; 24extern unsigned long max_mapnr;
@@ -1449,9 +1448,6 @@ int vmemmap_populate_basepages(struct page *start_page,
1449int vmemmap_populate(struct page *start_page, unsigned long pages, int node); 1448int vmemmap_populate(struct page *start_page, unsigned long pages, int node);
1450void vmemmap_populate_print_last(void); 1449void vmemmap_populate_print_last(void);
1451 1450
1452extern int account_locked_memory(struct mm_struct *mm, struct rlimit *rlim,
1453 size_t size);
1454extern void refund_locked_memory(struct mm_struct *mm, size_t size);
1455 1451
1456enum mf_flags { 1452enum mf_flags {
1457 MF_COUNT_INCREASED = 1 << 0, 1453 MF_COUNT_INCREASED = 1 << 0,
diff --git a/include/linux/ptrace.h b/include/linux/ptrace.h
index e1fb60729979..4272521e29e9 100644
--- a/include/linux/ptrace.h
+++ b/include/linux/ptrace.h
@@ -345,18 +345,6 @@ static inline void user_single_step_siginfo(struct task_struct *tsk,
345#define arch_ptrace_stop(code, info) do { } while (0) 345#define arch_ptrace_stop(code, info) do { } while (0)
346#endif 346#endif
347 347
348#ifndef arch_ptrace_untrace
349/*
350 * Do machine-specific work before untracing child.
351 *
352 * This is called for a normal detach as well as from ptrace_exit()
353 * when the tracing task dies.
354 *
355 * Called with write_lock(&tasklist_lock) held.
356 */
357#define arch_ptrace_untrace(task) do { } while (0)
358#endif
359
360extern int task_current_syscall(struct task_struct *target, long *callno, 348extern int task_current_syscall(struct task_struct *target, long *callno,
361 unsigned long args[6], unsigned int maxargs, 349 unsigned long args[6], unsigned int maxargs,
362 unsigned long *sp, unsigned long *pc); 350 unsigned long *sp, unsigned long *pc);
diff --git a/include/linux/sched.h b/include/linux/sched.h
index dad7f668ebf7..e0447c64af6a 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -99,7 +99,6 @@ struct futex_pi_state;
99struct robust_list_head; 99struct robust_list_head;
100struct bio_list; 100struct bio_list;
101struct fs_struct; 101struct fs_struct;
102struct bts_context;
103struct perf_event_context; 102struct perf_event_context;
104 103
105/* 104/*
@@ -1272,12 +1271,6 @@ struct task_struct {
1272 struct list_head ptraced; 1271 struct list_head ptraced;
1273 struct list_head ptrace_entry; 1272 struct list_head ptrace_entry;
1274 1273
1275 /*
1276 * This is the tracer handle for the ptrace BTS extension.
1277 * This field actually belongs to the ptracer task.
1278 */
1279 struct bts_context *bts;
1280
1281 /* PID/PID hash table linkage. */ 1274 /* PID/PID hash table linkage. */
1282 struct pid_link pids[PIDTYPE_MAX]; 1275 struct pid_link pids[PIDTYPE_MAX];
1283 struct list_head thread_group; 1276 struct list_head thread_group;
@@ -2123,10 +2116,8 @@ extern void set_task_comm(struct task_struct *tsk, char *from);
2123extern char *get_task_comm(char *to, struct task_struct *tsk); 2116extern char *get_task_comm(char *to, struct task_struct *tsk);
2124 2117
2125#ifdef CONFIG_SMP 2118#ifdef CONFIG_SMP
2126extern void wait_task_context_switch(struct task_struct *p);
2127extern unsigned long wait_task_inactive(struct task_struct *, long match_state); 2119extern unsigned long wait_task_inactive(struct task_struct *, long match_state);
2128#else 2120#else
2129static inline void wait_task_context_switch(struct task_struct *p) {}
2130static inline unsigned long wait_task_inactive(struct task_struct *p, 2121static inline unsigned long wait_task_inactive(struct task_struct *p,
2131 long match_state) 2122 long match_state)
2132{ 2123{
diff --git a/kernel/fork.c b/kernel/fork.c
index 4799c5f0e6d0..d67f1dbfbe03 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -1108,9 +1108,6 @@ static struct task_struct *copy_process(unsigned long clone_flags,
1108 p->memcg_batch.do_batch = 0; 1108 p->memcg_batch.do_batch = 0;
1109 p->memcg_batch.memcg = NULL; 1109 p->memcg_batch.memcg = NULL;
1110#endif 1110#endif
1111
1112 p->bts = NULL;
1113
1114 p->stack_start = stack_start; 1111 p->stack_start = stack_start;
1115 1112
1116 /* Perform scheduler related setup. Assign this task to a CPU. */ 1113 /* Perform scheduler related setup. Assign this task to a CPU. */
diff --git a/kernel/ptrace.c b/kernel/ptrace.c
index 42ad8ae729a0..9fb51237b18c 100644
--- a/kernel/ptrace.c
+++ b/kernel/ptrace.c
@@ -76,7 +76,6 @@ void __ptrace_unlink(struct task_struct *child)
76 child->parent = child->real_parent; 76 child->parent = child->real_parent;
77 list_del_init(&child->ptrace_entry); 77 list_del_init(&child->ptrace_entry);
78 78
79 arch_ptrace_untrace(child);
80 if (task_is_traced(child)) 79 if (task_is_traced(child))
81 ptrace_untrace(child); 80 ptrace_untrace(child);
82} 81}
diff --git a/kernel/sched.c b/kernel/sched.c
index 9ab3cd7858d3..117b7cad31b3 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -2077,49 +2077,6 @@ migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req)
2077} 2077}
2078 2078
2079/* 2079/*
2080 * wait_task_context_switch - wait for a thread to complete at least one
2081 * context switch.
2082 *
2083 * @p must not be current.
2084 */
2085void wait_task_context_switch(struct task_struct *p)
2086{
2087 unsigned long nvcsw, nivcsw, flags;
2088 int running;
2089 struct rq *rq;
2090
2091 nvcsw = p->nvcsw;
2092 nivcsw = p->nivcsw;
2093 for (;;) {
2094 /*
2095 * The runqueue is assigned before the actual context
2096 * switch. We need to take the runqueue lock.
2097 *
2098 * We could check initially without the lock but it is
2099 * very likely that we need to take the lock in every
2100 * iteration.
2101 */
2102 rq = task_rq_lock(p, &flags);
2103 running = task_running(rq, p);
2104 task_rq_unlock(rq, &flags);
2105
2106 if (likely(!running))
2107 break;
2108 /*
2109 * The switch count is incremented before the actual
2110 * context switch. We thus wait for two switches to be
2111 * sure at least one completed.
2112 */
2113 if ((p->nvcsw - nvcsw) > 1)
2114 break;
2115 if ((p->nivcsw - nivcsw) > 1)
2116 break;
2117
2118 cpu_relax();
2119 }
2120}
2121
2122/*
2123 * wait_task_inactive - wait for a thread to unschedule. 2080 * wait_task_inactive - wait for a thread to unschedule.
2124 * 2081 *
2125 * If @match_state is nonzero, it's the @p->state value just checked and 2082 * If @match_state is nonzero, it's the @p->state value just checked and
diff --git a/kernel/trace/Kconfig b/kernel/trace/Kconfig
index 13e13d428cd3..8b1797c4545b 100644
--- a/kernel/trace/Kconfig
+++ b/kernel/trace/Kconfig
@@ -44,9 +44,6 @@ config HAVE_FTRACE_MCOUNT_RECORD
44 help 44 help
45 See Documentation/trace/ftrace-design.txt 45 See Documentation/trace/ftrace-design.txt
46 46
47config HAVE_HW_BRANCH_TRACER
48 bool
49
50config HAVE_SYSCALL_TRACEPOINTS 47config HAVE_SYSCALL_TRACEPOINTS
51 bool 48 bool
52 help 49 help
@@ -374,14 +371,6 @@ config STACK_TRACER
374 371
375 Say N if unsure. 372 Say N if unsure.
376 373
377config HW_BRANCH_TRACER
378 depends on HAVE_HW_BRANCH_TRACER
379 bool "Trace hw branches"
380 select GENERIC_TRACER
381 help
382 This tracer records all branches on the system in a circular
383 buffer, giving access to the last N branches for each cpu.
384
385config KMEMTRACE 374config KMEMTRACE
386 bool "Trace SLAB allocations" 375 bool "Trace SLAB allocations"
387 select GENERIC_TRACER 376 select GENERIC_TRACER
diff --git a/kernel/trace/Makefile b/kernel/trace/Makefile
index 78edc6490038..ffb1a5b0550e 100644
--- a/kernel/trace/Makefile
+++ b/kernel/trace/Makefile
@@ -41,7 +41,6 @@ obj-$(CONFIG_MMIOTRACE) += trace_mmiotrace.o
41obj-$(CONFIG_BOOT_TRACER) += trace_boot.o 41obj-$(CONFIG_BOOT_TRACER) += trace_boot.o
42obj-$(CONFIG_FUNCTION_GRAPH_TRACER) += trace_functions_graph.o 42obj-$(CONFIG_FUNCTION_GRAPH_TRACER) += trace_functions_graph.o
43obj-$(CONFIG_TRACE_BRANCH_PROFILING) += trace_branch.o 43obj-$(CONFIG_TRACE_BRANCH_PROFILING) += trace_branch.o
44obj-$(CONFIG_HW_BRANCH_TRACER) += trace_hw_branches.o
45obj-$(CONFIG_KMEMTRACE) += kmemtrace.o 44obj-$(CONFIG_KMEMTRACE) += kmemtrace.o
46obj-$(CONFIG_WORKQUEUE_TRACER) += trace_workqueue.o 45obj-$(CONFIG_WORKQUEUE_TRACER) += trace_workqueue.o
47obj-$(CONFIG_BLK_DEV_IO_TRACE) += blktrace.o 46obj-$(CONFIG_BLK_DEV_IO_TRACE) += blktrace.o
diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
index 2825ef2c0b15..bec2c973ff0c 100644
--- a/kernel/trace/trace.h
+++ b/kernel/trace/trace.h
@@ -34,7 +34,6 @@ enum trace_type {
34 TRACE_GRAPH_RET, 34 TRACE_GRAPH_RET,
35 TRACE_GRAPH_ENT, 35 TRACE_GRAPH_ENT,
36 TRACE_USER_STACK, 36 TRACE_USER_STACK,
37 TRACE_HW_BRANCHES,
38 TRACE_KMEM_ALLOC, 37 TRACE_KMEM_ALLOC,
39 TRACE_KMEM_FREE, 38 TRACE_KMEM_FREE,
40 TRACE_BLK, 39 TRACE_BLK,
@@ -229,7 +228,6 @@ extern void __ftrace_bad_type(void);
229 TRACE_GRAPH_ENT); \ 228 TRACE_GRAPH_ENT); \
230 IF_ASSIGN(var, ent, struct ftrace_graph_ret_entry, \ 229 IF_ASSIGN(var, ent, struct ftrace_graph_ret_entry, \
231 TRACE_GRAPH_RET); \ 230 TRACE_GRAPH_RET); \
232 IF_ASSIGN(var, ent, struct hw_branch_entry, TRACE_HW_BRANCHES);\
233 IF_ASSIGN(var, ent, struct kmemtrace_alloc_entry, \ 231 IF_ASSIGN(var, ent, struct kmemtrace_alloc_entry, \
234 TRACE_KMEM_ALLOC); \ 232 TRACE_KMEM_ALLOC); \
235 IF_ASSIGN(var, ent, struct kmemtrace_free_entry, \ 233 IF_ASSIGN(var, ent, struct kmemtrace_free_entry, \
@@ -467,8 +465,6 @@ extern int trace_selftest_startup_sysprof(struct tracer *trace,
467 struct trace_array *tr); 465 struct trace_array *tr);
468extern int trace_selftest_startup_branch(struct tracer *trace, 466extern int trace_selftest_startup_branch(struct tracer *trace,
469 struct trace_array *tr); 467 struct trace_array *tr);
470extern int trace_selftest_startup_hw_branches(struct tracer *trace,
471 struct trace_array *tr);
472extern int trace_selftest_startup_ksym(struct tracer *trace, 468extern int trace_selftest_startup_ksym(struct tracer *trace,
473 struct trace_array *tr); 469 struct trace_array *tr);
474#endif /* CONFIG_FTRACE_STARTUP_TEST */ 470#endif /* CONFIG_FTRACE_STARTUP_TEST */
diff --git a/kernel/trace/trace_entries.h b/kernel/trace/trace_entries.h
index c16a08f399df..dc008c1240da 100644
--- a/kernel/trace/trace_entries.h
+++ b/kernel/trace/trace_entries.h
@@ -318,18 +318,6 @@ FTRACE_ENTRY(branch, trace_branch,
318 __entry->func, __entry->file, __entry->correct) 318 __entry->func, __entry->file, __entry->correct)
319); 319);
320 320
321FTRACE_ENTRY(hw_branch, hw_branch_entry,
322
323 TRACE_HW_BRANCHES,
324
325 F_STRUCT(
326 __field( u64, from )
327 __field( u64, to )
328 ),
329
330 F_printk("from: %llx to: %llx", __entry->from, __entry->to)
331);
332
333FTRACE_ENTRY(kmem_alloc, kmemtrace_alloc_entry, 321FTRACE_ENTRY(kmem_alloc, kmemtrace_alloc_entry,
334 322
335 TRACE_KMEM_ALLOC, 323 TRACE_KMEM_ALLOC,
diff --git a/kernel/trace/trace_hw_branches.c b/kernel/trace/trace_hw_branches.c
deleted file mode 100644
index 7b97000745f5..000000000000
--- a/kernel/trace/trace_hw_branches.c
+++ /dev/null
@@ -1,312 +0,0 @@
1/*
2 * h/w branch tracer for x86 based on BTS
3 *
4 * Copyright (C) 2008-2009 Intel Corporation.
5 * Markus Metzger <markus.t.metzger@gmail.com>, 2008-2009
6 */
7#include <linux/kallsyms.h>
8#include <linux/debugfs.h>
9#include <linux/ftrace.h>
10#include <linux/module.h>
11#include <linux/cpu.h>
12#include <linux/smp.h>
13#include <linux/fs.h>
14
15#include <asm/ds.h>
16
17#include "trace_output.h"
18#include "trace.h"
19
20
21#define BTS_BUFFER_SIZE (1 << 13)
22
23static DEFINE_PER_CPU(struct bts_tracer *, hwb_tracer);
24static DEFINE_PER_CPU(unsigned char[BTS_BUFFER_SIZE], hwb_buffer);
25
26#define this_tracer per_cpu(hwb_tracer, smp_processor_id())
27
28static int trace_hw_branches_enabled __read_mostly;
29static int trace_hw_branches_suspended __read_mostly;
30static struct trace_array *hw_branch_trace __read_mostly;
31
32
33static void bts_trace_init_cpu(int cpu)
34{
35 per_cpu(hwb_tracer, cpu) =
36 ds_request_bts_cpu(cpu, per_cpu(hwb_buffer, cpu),
37 BTS_BUFFER_SIZE, NULL, (size_t)-1,
38 BTS_KERNEL);
39
40 if (IS_ERR(per_cpu(hwb_tracer, cpu)))
41 per_cpu(hwb_tracer, cpu) = NULL;
42}
43
44static int bts_trace_init(struct trace_array *tr)
45{
46 int cpu;
47
48 hw_branch_trace = tr;
49 trace_hw_branches_enabled = 0;
50
51 get_online_cpus();
52 for_each_online_cpu(cpu) {
53 bts_trace_init_cpu(cpu);
54
55 if (likely(per_cpu(hwb_tracer, cpu)))
56 trace_hw_branches_enabled = 1;
57 }
58 trace_hw_branches_suspended = 0;
59 put_online_cpus();
60
61 /* If we could not enable tracing on a single cpu, we fail. */
62 return trace_hw_branches_enabled ? 0 : -EOPNOTSUPP;
63}
64
65static void bts_trace_reset(struct trace_array *tr)
66{
67 int cpu;
68
69 get_online_cpus();
70 for_each_online_cpu(cpu) {
71 if (likely(per_cpu(hwb_tracer, cpu))) {
72 ds_release_bts(per_cpu(hwb_tracer, cpu));
73 per_cpu(hwb_tracer, cpu) = NULL;
74 }
75 }
76 trace_hw_branches_enabled = 0;
77 trace_hw_branches_suspended = 0;
78 put_online_cpus();
79}
80
81static void bts_trace_start(struct trace_array *tr)
82{
83 int cpu;
84
85 get_online_cpus();
86 for_each_online_cpu(cpu)
87 if (likely(per_cpu(hwb_tracer, cpu)))
88 ds_resume_bts(per_cpu(hwb_tracer, cpu));
89 trace_hw_branches_suspended = 0;
90 put_online_cpus();
91}
92
93static void bts_trace_stop(struct trace_array *tr)
94{
95 int cpu;
96
97 get_online_cpus();
98 for_each_online_cpu(cpu)
99 if (likely(per_cpu(hwb_tracer, cpu)))
100 ds_suspend_bts(per_cpu(hwb_tracer, cpu));
101 trace_hw_branches_suspended = 1;
102 put_online_cpus();
103}
104
105static int __cpuinit bts_hotcpu_handler(struct notifier_block *nfb,
106 unsigned long action, void *hcpu)
107{
108 int cpu = (long)hcpu;
109
110 switch (action) {
111 case CPU_ONLINE:
112 case CPU_DOWN_FAILED:
113 /* The notification is sent with interrupts enabled. */
114 if (trace_hw_branches_enabled) {
115 bts_trace_init_cpu(cpu);
116
117 if (trace_hw_branches_suspended &&
118 likely(per_cpu(hwb_tracer, cpu)))
119 ds_suspend_bts(per_cpu(hwb_tracer, cpu));
120 }
121 break;
122
123 case CPU_DOWN_PREPARE:
124 /* The notification is sent with interrupts enabled. */
125 if (likely(per_cpu(hwb_tracer, cpu))) {
126 ds_release_bts(per_cpu(hwb_tracer, cpu));
127 per_cpu(hwb_tracer, cpu) = NULL;
128 }
129 }
130
131 return NOTIFY_DONE;
132}
133
134static struct notifier_block bts_hotcpu_notifier __cpuinitdata = {
135 .notifier_call = bts_hotcpu_handler
136};
137
138static void bts_trace_print_header(struct seq_file *m)
139{
140 seq_puts(m, "# CPU# TO <- FROM\n");
141}
142
143static enum print_line_t bts_trace_print_line(struct trace_iterator *iter)
144{
145 unsigned long symflags = TRACE_ITER_SYM_OFFSET;
146 struct trace_entry *entry = iter->ent;
147 struct trace_seq *seq = &iter->seq;
148 struct hw_branch_entry *it;
149
150 trace_assign_type(it, entry);
151
152 if (entry->type == TRACE_HW_BRANCHES) {
153 if (trace_seq_printf(seq, "%4d ", iter->cpu) &&
154 seq_print_ip_sym(seq, it->to, symflags) &&
155 trace_seq_printf(seq, "\t <- ") &&
156 seq_print_ip_sym(seq, it->from, symflags) &&
157 trace_seq_printf(seq, "\n"))
158 return TRACE_TYPE_HANDLED;
159 return TRACE_TYPE_PARTIAL_LINE;
160 }
161 return TRACE_TYPE_UNHANDLED;
162}
163
164void trace_hw_branch(u64 from, u64 to)
165{
166 struct ftrace_event_call *call = &event_hw_branch;
167 struct trace_array *tr = hw_branch_trace;
168 struct ring_buffer_event *event;
169 struct ring_buffer *buf;
170 struct hw_branch_entry *entry;
171 unsigned long irq1;
172 int cpu;
173
174 if (unlikely(!tr))
175 return;
176
177 if (unlikely(!trace_hw_branches_enabled))
178 return;
179
180 local_irq_save(irq1);
181 cpu = raw_smp_processor_id();
182 if (atomic_inc_return(&tr->data[cpu]->disabled) != 1)
183 goto out;
184
185 buf = tr->buffer;
186 event = trace_buffer_lock_reserve(buf, TRACE_HW_BRANCHES,
187 sizeof(*entry), 0, 0);
188 if (!event)
189 goto out;
190 entry = ring_buffer_event_data(event);
191 tracing_generic_entry_update(&entry->ent, 0, from);
192 entry->ent.type = TRACE_HW_BRANCHES;
193 entry->from = from;
194 entry->to = to;
195 if (!filter_check_discard(call, entry, buf, event))
196 trace_buffer_unlock_commit(buf, event, 0, 0);
197
198 out:
199 atomic_dec(&tr->data[cpu]->disabled);
200 local_irq_restore(irq1);
201}
202
203static void trace_bts_at(const struct bts_trace *trace, void *at)
204{
205 struct bts_struct bts;
206 int err = 0;
207
208 WARN_ON_ONCE(!trace->read);
209 if (!trace->read)
210 return;
211
212 err = trace->read(this_tracer, at, &bts);
213 if (err < 0)
214 return;
215
216 switch (bts.qualifier) {
217 case BTS_BRANCH:
218 trace_hw_branch(bts.variant.lbr.from, bts.variant.lbr.to);
219 break;
220 }
221}
222
223/*
224 * Collect the trace on the current cpu and write it into the ftrace buffer.
225 *
226 * pre: tracing must be suspended on the current cpu
227 */
228static void trace_bts_cpu(void *arg)
229{
230 struct trace_array *tr = (struct trace_array *)arg;
231 const struct bts_trace *trace;
232 unsigned char *at;
233
234 if (unlikely(!tr))
235 return;
236
237 if (unlikely(atomic_read(&tr->data[raw_smp_processor_id()]->disabled)))
238 return;
239
240 if (unlikely(!this_tracer))
241 return;
242
243 trace = ds_read_bts(this_tracer);
244 if (!trace)
245 return;
246
247 for (at = trace->ds.top; (void *)at < trace->ds.end;
248 at += trace->ds.size)
249 trace_bts_at(trace, at);
250
251 for (at = trace->ds.begin; (void *)at < trace->ds.top;
252 at += trace->ds.size)
253 trace_bts_at(trace, at);
254}
255
256static void trace_bts_prepare(struct trace_iterator *iter)
257{
258 int cpu;
259
260 get_online_cpus();
261 for_each_online_cpu(cpu)
262 if (likely(per_cpu(hwb_tracer, cpu)))
263 ds_suspend_bts(per_cpu(hwb_tracer, cpu));
264 /*
265 * We need to collect the trace on the respective cpu since ftrace
266 * implicitly adds the record for the current cpu.
267 * Once that is more flexible, we could collect the data from any cpu.
268 */
269 on_each_cpu(trace_bts_cpu, iter->tr, 1);
270
271 for_each_online_cpu(cpu)
272 if (likely(per_cpu(hwb_tracer, cpu)))
273 ds_resume_bts(per_cpu(hwb_tracer, cpu));
274 put_online_cpus();
275}
276
277static void trace_bts_close(struct trace_iterator *iter)
278{
279 tracing_reset_online_cpus(iter->tr);
280}
281
282void trace_hw_branch_oops(void)
283{
284 if (this_tracer) {
285 ds_suspend_bts_noirq(this_tracer);
286 trace_bts_cpu(hw_branch_trace);
287 ds_resume_bts_noirq(this_tracer);
288 }
289}
290
291struct tracer bts_tracer __read_mostly =
292{
293 .name = "hw-branch-tracer",
294 .init = bts_trace_init,
295 .reset = bts_trace_reset,
296 .print_header = bts_trace_print_header,
297 .print_line = bts_trace_print_line,
298 .start = bts_trace_start,
299 .stop = bts_trace_stop,
300 .open = trace_bts_prepare,
301 .close = trace_bts_close,
302#ifdef CONFIG_FTRACE_SELFTEST
303 .selftest = trace_selftest_startup_hw_branches,
304#endif /* CONFIG_FTRACE_SELFTEST */
305};
306
307__init static int init_bts_trace(void)
308{
309 register_hotcpu_notifier(&bts_hotcpu_notifier);
310 return register_tracer(&bts_tracer);
311}
312device_initcall(init_bts_trace);
diff --git a/kernel/trace/trace_selftest.c b/kernel/trace/trace_selftest.c
index 280fea470d67..a7084e7c0427 100644
--- a/kernel/trace/trace_selftest.c
+++ b/kernel/trace/trace_selftest.c
@@ -16,7 +16,6 @@ static inline int trace_valid_entry(struct trace_entry *entry)
16 case TRACE_BRANCH: 16 case TRACE_BRANCH:
17 case TRACE_GRAPH_ENT: 17 case TRACE_GRAPH_ENT:
18 case TRACE_GRAPH_RET: 18 case TRACE_GRAPH_RET:
19 case TRACE_HW_BRANCHES:
20 case TRACE_KSYM: 19 case TRACE_KSYM:
21 return 1; 20 return 1;
22 } 21 }
@@ -754,62 +753,6 @@ trace_selftest_startup_branch(struct tracer *trace, struct trace_array *tr)
754} 753}
755#endif /* CONFIG_BRANCH_TRACER */ 754#endif /* CONFIG_BRANCH_TRACER */
756 755
757#ifdef CONFIG_HW_BRANCH_TRACER
758int
759trace_selftest_startup_hw_branches(struct tracer *trace,
760 struct trace_array *tr)
761{
762 struct trace_iterator *iter;
763 struct tracer tracer;
764 unsigned long count;
765 int ret;
766
767 if (!trace->open) {
768 printk(KERN_CONT "missing open function...");
769 return -1;
770 }
771
772 ret = tracer_init(trace, tr);
773 if (ret) {
774 warn_failed_init_tracer(trace, ret);
775 return ret;
776 }
777
778 /*
779 * The hw-branch tracer needs to collect the trace from the various
780 * cpu trace buffers - before tracing is stopped.
781 */
782 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
783 if (!iter)
784 return -ENOMEM;
785
786 memcpy(&tracer, trace, sizeof(tracer));
787
788 iter->trace = &tracer;
789 iter->tr = tr;
790 iter->pos = -1;
791 mutex_init(&iter->mutex);
792
793 trace->open(iter);
794
795 mutex_destroy(&iter->mutex);
796 kfree(iter);
797
798 tracing_stop();
799
800 ret = trace_test_buffer(tr, &count);
801 trace->reset(tr);
802 tracing_start();
803
804 if (!ret && !count) {
805 printk(KERN_CONT "no entries found..");
806 ret = -1;
807 }
808
809 return ret;
810}
811#endif /* CONFIG_HW_BRANCH_TRACER */
812
813#ifdef CONFIG_KSYM_TRACER 756#ifdef CONFIG_KSYM_TRACER
814static int ksym_selftest_dummy; 757static int ksym_selftest_dummy;
815 758
diff --git a/mm/mlock.c b/mm/mlock.c
index 8f4e2dfceec1..3f82720e0515 100644
--- a/mm/mlock.c
+++ b/mm/mlock.c
@@ -607,44 +607,3 @@ void user_shm_unlock(size_t size, struct user_struct *user)
607 spin_unlock(&shmlock_user_lock); 607 spin_unlock(&shmlock_user_lock);
608 free_uid(user); 608 free_uid(user);
609} 609}
610
611int account_locked_memory(struct mm_struct *mm, struct rlimit *rlim,
612 size_t size)
613{
614 unsigned long lim, vm, pgsz;
615 int error = -ENOMEM;
616
617 pgsz = PAGE_ALIGN(size) >> PAGE_SHIFT;
618
619 down_write(&mm->mmap_sem);
620
621 lim = ACCESS_ONCE(rlim[RLIMIT_AS].rlim_cur) >> PAGE_SHIFT;
622 vm = mm->total_vm + pgsz;
623 if (lim < vm)
624 goto out;
625
626 lim = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur) >> PAGE_SHIFT;
627 vm = mm->locked_vm + pgsz;
628 if (lim < vm)
629 goto out;
630
631 mm->total_vm += pgsz;
632 mm->locked_vm += pgsz;
633
634 error = 0;
635 out:
636 up_write(&mm->mmap_sem);
637 return error;
638}
639
640void refund_locked_memory(struct mm_struct *mm, size_t size)
641{
642 unsigned long pgsz = PAGE_ALIGN(size) >> PAGE_SHIFT;
643
644 down_write(&mm->mmap_sem);
645
646 mm->total_vm -= pgsz;
647 mm->locked_vm -= pgsz;
648
649 up_write(&mm->mmap_sem);
650}
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-05 22:28:59 -0500