aboutsummaryrefslogtreecommitdiffstats
path: root/arch
diff options
context:
space:
mode:
Diffstat (limited to 'arch')
-rw-r--r--arch/x86/Kconfig.debug9
-rw-r--r--arch/x86/include/asm/ds.h82
-rw-r--r--arch/x86/include/asm/processor.h35
-rw-r--r--arch/x86/include/asm/ptrace.h9
-rw-r--r--arch/x86/include/asm/tlbflush.h8
-rw-r--r--arch/x86/kernel/Makefile1
-rw-r--r--arch/x86/kernel/ds.c921
-rw-r--r--arch/x86/kernel/ds_selftest.c408
-rw-r--r--arch/x86/kernel/ds_selftest.h15
-rw-r--r--arch/x86/kernel/entry_64.S19
-rw-r--r--arch/x86/kernel/process.c5
-rw-r--r--arch/x86/kernel/process_32.c3
-rw-r--r--arch/x86/kernel/process_64.c3
-rw-r--r--arch/x86/kernel/ptrace.c284
-rw-r--r--arch/x86/kernel/stacktrace.c2
-rw-r--r--arch/x86/mm/kmmio.c104
-rw-r--r--arch/x86/mm/mmio-mod.c2
17 files changed, 1444 insertions, 466 deletions
diff --git a/arch/x86/Kconfig.debug b/arch/x86/Kconfig.debug
index 33fac6bbe1c2..d105f29bb6bb 100644
--- a/arch/x86/Kconfig.debug
+++ b/arch/x86/Kconfig.debug
@@ -174,6 +174,15 @@ 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
177config HAVE_MMIOTRACE_SUPPORT 186config HAVE_MMIOTRACE_SUPPORT
178 def_bool y 187 def_bool y
179 188
diff --git a/arch/x86/include/asm/ds.h b/arch/x86/include/asm/ds.h
index a8f672ba100c..70dac199b093 100644
--- a/arch/x86/include/asm/ds.h
+++ b/arch/x86/include/asm/ds.h
@@ -15,8 +15,8 @@
15 * - buffer allocation (memory accounting) 15 * - buffer allocation (memory accounting)
16 * 16 *
17 * 17 *
18 * Copyright (C) 2007-2008 Intel Corporation. 18 * Copyright (C) 2007-2009 Intel Corporation.
19 * Markus Metzger <markus.t.metzger@intel.com>, 2007-2008 19 * Markus Metzger <markus.t.metzger@intel.com>, 2007-2009
20 */ 20 */
21 21
22#ifndef _ASM_X86_DS_H 22#ifndef _ASM_X86_DS_H
@@ -83,8 +83,10 @@ enum ds_feature {
83 * The interrupt threshold is independent from the overflow callback 83 * The interrupt threshold is independent from the overflow callback
84 * to allow users to use their own overflow interrupt handling mechanism. 84 * to allow users to use their own overflow interrupt handling mechanism.
85 * 85 *
86 * task: the task to request recording for; 86 * The function might sleep.
87 * NULL for per-cpu recording on the current cpu 87 *
88 * task: the task to request recording for
89 * cpu: the cpu to request recording for
88 * base: the base pointer for the (non-pageable) buffer; 90 * base: the base pointer for the (non-pageable) buffer;
89 * size: the size of the provided buffer in bytes 91 * size: the size of the provided buffer in bytes
90 * ovfl: pointer to a function to be called on buffer overflow; 92 * ovfl: pointer to a function to be called on buffer overflow;
@@ -93,19 +95,28 @@ enum ds_feature {
93 * -1 if no interrupt threshold is requested. 95 * -1 if no interrupt threshold is requested.
94 * flags: a bit-mask of the above flags 96 * flags: a bit-mask of the above flags
95 */ 97 */
96extern struct bts_tracer *ds_request_bts(struct task_struct *task, 98extern struct bts_tracer *ds_request_bts_task(struct task_struct *task,
97 void *base, size_t size, 99 void *base, size_t size,
98 bts_ovfl_callback_t ovfl, 100 bts_ovfl_callback_t ovfl,
99 size_t th, unsigned int flags); 101 size_t th, unsigned int flags);
100extern struct pebs_tracer *ds_request_pebs(struct task_struct *task, 102extern struct bts_tracer *ds_request_bts_cpu(int cpu, void *base, size_t size,
101 void *base, size_t size, 103 bts_ovfl_callback_t ovfl,
102 pebs_ovfl_callback_t ovfl, 104 size_t th, unsigned int flags);
103 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);
104 113
105/* 114/*
106 * Release BTS or PEBS resources 115 * Release BTS or PEBS resources
107 * Suspend and resume BTS or PEBS tracing 116 * Suspend and resume BTS or PEBS tracing
108 * 117 *
118 * Must be called with irq's enabled.
119 *
109 * tracer: the tracer handle returned from ds_request_~() 120 * tracer: the tracer handle returned from ds_request_~()
110 */ 121 */
111extern void ds_release_bts(struct bts_tracer *tracer); 122extern void ds_release_bts(struct bts_tracer *tracer);
@@ -115,6 +126,28 @@ extern void ds_release_pebs(struct pebs_tracer *tracer);
115extern void ds_suspend_pebs(struct pebs_tracer *tracer); 126extern void ds_suspend_pebs(struct pebs_tracer *tracer);
116extern void ds_resume_pebs(struct pebs_tracer *tracer); 127extern void ds_resume_pebs(struct pebs_tracer *tracer);
117 128
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
118 151
119/* 152/*
120 * The raw DS buffer state as it is used for BTS and PEBS recording. 153 * The raw DS buffer state as it is used for BTS and PEBS recording.
@@ -170,9 +203,9 @@ struct bts_struct {
170 } lbr; 203 } lbr;
171 /* BTS_TASK_ARRIVES or BTS_TASK_DEPARTS */ 204 /* BTS_TASK_ARRIVES or BTS_TASK_DEPARTS */
172 struct { 205 struct {
173 __u64 jiffies; 206 __u64 clock;
174 pid_t pid; 207 pid_t pid;
175 } timestamp; 208 } event;
176 } variant; 209 } variant;
177}; 210};
178 211
@@ -201,8 +234,12 @@ struct bts_trace {
201struct pebs_trace { 234struct pebs_trace {
202 struct ds_trace ds; 235 struct ds_trace ds;
203 236
204 /* the PEBS reset value */ 237 /* the number of valid counters in the below array */
205 unsigned long long reset_value; 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];
206}; 243};
207 244
208 245
@@ -237,9 +274,11 @@ extern int ds_reset_pebs(struct pebs_tracer *tracer);
237 * Returns 0 on success; -Eerrno on error 274 * Returns 0 on success; -Eerrno on error
238 * 275 *
239 * tracer: the tracer handle returned from ds_request_pebs() 276 * tracer: the tracer handle returned from ds_request_pebs()
277 * counter: the index of the counter
240 * value: the new counter reset value 278 * value: the new counter reset value
241 */ 279 */
242extern int ds_set_pebs_reset(struct pebs_tracer *tracer, u64 value); 280extern int ds_set_pebs_reset(struct pebs_tracer *tracer,
281 unsigned int counter, u64 value);
243 282
244/* 283/*
245 * Initialization 284 * Initialization
@@ -252,21 +291,12 @@ extern void __cpuinit ds_init_intel(struct cpuinfo_x86 *);
252 */ 291 */
253extern void ds_switch_to(struct task_struct *prev, struct task_struct *next); 292extern void ds_switch_to(struct task_struct *prev, struct task_struct *next);
254 293
255/*
256 * Task clone/init and cleanup work
257 */
258extern void ds_copy_thread(struct task_struct *tsk, struct task_struct *father);
259extern void ds_exit_thread(struct task_struct *tsk);
260
261#else /* CONFIG_X86_DS */ 294#else /* CONFIG_X86_DS */
262 295
263struct cpuinfo_x86; 296struct cpuinfo_x86;
264static inline void __cpuinit ds_init_intel(struct cpuinfo_x86 *ignored) {} 297static inline void __cpuinit ds_init_intel(struct cpuinfo_x86 *ignored) {}
265static inline void ds_switch_to(struct task_struct *prev, 298static inline void ds_switch_to(struct task_struct *prev,
266 struct task_struct *next) {} 299 struct task_struct *next) {}
267static inline void ds_copy_thread(struct task_struct *tsk,
268 struct task_struct *father) {}
269static inline void ds_exit_thread(struct task_struct *tsk) {}
270 300
271#endif /* CONFIG_X86_DS */ 301#endif /* CONFIG_X86_DS */
272#endif /* _ASM_X86_DS_H */ 302#endif /* _ASM_X86_DS_H */
diff --git a/arch/x86/include/asm/processor.h b/arch/x86/include/asm/processor.h
index 87ede2f31bc7..c7768269b1cf 100644
--- a/arch/x86/include/asm/processor.h
+++ b/arch/x86/include/asm/processor.h
@@ -462,14 +462,8 @@ struct thread_struct {
462 unsigned io_bitmap_max; 462 unsigned io_bitmap_max;
463/* MSR_IA32_DEBUGCTLMSR value to switch in if TIF_DEBUGCTLMSR is set. */ 463/* MSR_IA32_DEBUGCTLMSR value to switch in if TIF_DEBUGCTLMSR is set. */
464 unsigned long debugctlmsr; 464 unsigned long debugctlmsr;
465#ifdef CONFIG_X86_DS 465 /* Debug Store context; see asm/ds.h */
466/* Debug Store context; see include/asm-x86/ds.h; goes into MSR_IA32_DS_AREA */
467 struct ds_context *ds_ctx; 466 struct ds_context *ds_ctx;
468#endif /* CONFIG_X86_DS */
469#ifdef CONFIG_X86_PTRACE_BTS
470/* the signal to send on a bts buffer overflow */
471 unsigned int bts_ovfl_signal;
472#endif /* CONFIG_X86_PTRACE_BTS */
473}; 467};
474 468
475static inline unsigned long native_get_debugreg(int regno) 469static inline unsigned long native_get_debugreg(int regno)
@@ -797,6 +791,21 @@ static inline unsigned long get_debugctlmsr(void)
797 return debugctlmsr; 791 return debugctlmsr;
798} 792}
799 793
794static inline unsigned long get_debugctlmsr_on_cpu(int cpu)
795{
796 u64 debugctlmsr = 0;
797 u32 val1, val2;
798
799#ifndef CONFIG_X86_DEBUGCTLMSR
800 if (boot_cpu_data.x86 < 6)
801 return 0;
802#endif
803 rdmsr_on_cpu(cpu, MSR_IA32_DEBUGCTLMSR, &val1, &val2);
804 debugctlmsr = val1 | ((u64)val2 << 32);
805
806 return debugctlmsr;
807}
808
800static inline void update_debugctlmsr(unsigned long debugctlmsr) 809static inline void update_debugctlmsr(unsigned long debugctlmsr)
801{ 810{
802#ifndef CONFIG_X86_DEBUGCTLMSR 811#ifndef CONFIG_X86_DEBUGCTLMSR
@@ -806,6 +815,18 @@ static inline void update_debugctlmsr(unsigned long debugctlmsr)
806 wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr); 815 wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
807} 816}
808 817
818static inline void update_debugctlmsr_on_cpu(int cpu,
819 unsigned long debugctlmsr)
820{
821#ifndef CONFIG_X86_DEBUGCTLMSR
822 if (boot_cpu_data.x86 < 6)
823 return;
824#endif
825 wrmsr_on_cpu(cpu, MSR_IA32_DEBUGCTLMSR,
826 (u32)((u64)debugctlmsr),
827 (u32)((u64)debugctlmsr >> 32));
828}
829
809/* 830/*
810 * from system description table in BIOS. Mostly for MCA use, but 831 * from system description table in BIOS. Mostly for MCA use, but
811 * others may find it useful: 832 * others may find it useful:
diff --git a/arch/x86/include/asm/ptrace.h b/arch/x86/include/asm/ptrace.h
index 624f133943ed..0f0d908349aa 100644
--- a/arch/x86/include/asm/ptrace.h
+++ b/arch/x86/include/asm/ptrace.h
@@ -236,12 +236,11 @@ extern int do_get_thread_area(struct task_struct *p, int idx,
236extern int do_set_thread_area(struct task_struct *p, int idx, 236extern int do_set_thread_area(struct task_struct *p, int idx,
237 struct user_desc __user *info, int can_allocate); 237 struct user_desc __user *info, int can_allocate);
238 238
239extern void x86_ptrace_untrace(struct task_struct *); 239#ifdef CONFIG_X86_PTRACE_BTS
240extern void x86_ptrace_fork(struct task_struct *child, 240extern void ptrace_bts_untrace(struct task_struct *tsk);
241 unsigned long clone_flags);
242 241
243#define arch_ptrace_untrace(tsk) x86_ptrace_untrace(tsk) 242#define arch_ptrace_untrace(tsk) ptrace_bts_untrace(tsk)
244#define arch_ptrace_fork(child, flags) x86_ptrace_fork(child, flags) 243#endif /* CONFIG_X86_PTRACE_BTS */
245 244
246#endif /* __KERNEL__ */ 245#endif /* __KERNEL__ */
247 246
diff --git a/arch/x86/include/asm/tlbflush.h b/arch/x86/include/asm/tlbflush.h
index 16a5c84b0329..a5ecc9c33e92 100644
--- a/arch/x86/include/asm/tlbflush.h
+++ b/arch/x86/include/asm/tlbflush.h
@@ -17,7 +17,7 @@
17 17
18static inline void __native_flush_tlb(void) 18static inline void __native_flush_tlb(void)
19{ 19{
20 write_cr3(read_cr3()); 20 native_write_cr3(native_read_cr3());
21} 21}
22 22
23static inline void __native_flush_tlb_global(void) 23static inline void __native_flush_tlb_global(void)
@@ -32,11 +32,11 @@ static inline void __native_flush_tlb_global(void)
32 */ 32 */
33 raw_local_irq_save(flags); 33 raw_local_irq_save(flags);
34 34
35 cr4 = read_cr4(); 35 cr4 = native_read_cr4();
36 /* clear PGE */ 36 /* clear PGE */
37 write_cr4(cr4 & ~X86_CR4_PGE); 37 native_write_cr4(cr4 & ~X86_CR4_PGE);
38 /* write old PGE again and flush TLBs */ 38 /* write old PGE again and flush TLBs */
39 write_cr4(cr4); 39 native_write_cr4(cr4);
40 40
41 raw_local_irq_restore(flags); 41 raw_local_irq_restore(flags);
42} 42}
diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile
index 235f5927bb97..4f78bd682125 100644
--- a/arch/x86/kernel/Makefile
+++ b/arch/x86/kernel/Makefile
@@ -44,6 +44,7 @@ obj-y += process.o
44obj-y += i387.o xsave.o 44obj-y += i387.o xsave.o
45obj-y += ptrace.o 45obj-y += ptrace.o
46obj-$(CONFIG_X86_DS) += ds.o 46obj-$(CONFIG_X86_DS) += ds.o
47obj-$(CONFIG_X86_DS_SELFTEST) += ds_selftest.o
47obj-$(CONFIG_X86_32) += tls.o 48obj-$(CONFIG_X86_32) += tls.o
48obj-$(CONFIG_IA32_EMULATION) += tls.o 49obj-$(CONFIG_IA32_EMULATION) += tls.o
49obj-y += step.o 50obj-y += step.o
diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c
index 87b67e3a765a..48bfe1386038 100644
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -19,45 +19,61 @@
19 * Markus Metzger <markus.t.metzger@intel.com>, 2007-2009 19 * Markus Metzger <markus.t.metzger@intel.com>, 2007-2009
20 */ 20 */
21 21
22 22#include <linux/kernel.h>
23#include <asm/ds.h>
24
25#include <linux/errno.h>
26#include <linux/string.h> 23#include <linux/string.h>
27#include <linux/slab.h> 24#include <linux/errno.h>
28#include <linux/sched.h> 25#include <linux/sched.h>
26#include <linux/slab.h>
29#include <linux/mm.h> 27#include <linux/mm.h>
30#include <linux/kernel.h> 28#include <linux/trace_clock.h>
29
30#include <asm/ds.h>
31 31
32#include "ds_selftest.h"
32 33
33/* 34/*
34 * The configuration for a particular DS hardware implementation. 35 * The configuration for a particular DS hardware implementation:
35 */ 36 */
36struct ds_configuration { 37struct ds_configuration {
37 /* the name of the configuration */ 38 /* The name of the configuration: */
38 const char *name; 39 const char *name;
39 /* the size of one pointer-typed field in the DS structure and 40
40 in the BTS and PEBS buffers in bytes; 41 /* The size of pointer-typed fields in DS, BTS, and PEBS: */
41 this covers the first 8 DS fields related to buffer management. */ 42 unsigned char sizeof_ptr_field;
42 unsigned char sizeof_field; 43
43 /* the size of a BTS/PEBS record in bytes */ 44 /* The size of a BTS/PEBS record in bytes: */
44 unsigned char sizeof_rec[2]; 45 unsigned char sizeof_rec[2];
45 /* a series of bit-masks to control various features indexed 46
46 * by enum ds_feature */ 47 /* The number of pebs counter reset values in the DS structure. */
47 unsigned long ctl[dsf_ctl_max]; 48 unsigned char nr_counter_reset;
49
50 /* Control bit-masks indexed by enum ds_feature: */
51 unsigned long ctl[dsf_ctl_max];
48}; 52};
49static DEFINE_PER_CPU(struct ds_configuration, ds_cfg_array); 53static struct ds_configuration ds_cfg __read_mostly;
54
55
56/* Maximal size of a DS configuration: */
57#define MAX_SIZEOF_DS 0x80
50 58
51#define ds_cfg per_cpu(ds_cfg_array, smp_processor_id()) 59/* Maximal size of a BTS record: */
60#define MAX_SIZEOF_BTS (3 * 8)
52 61
53#define MAX_SIZEOF_DS (12 * 8) /* maximal size of a DS configuration */ 62/* BTS and PEBS buffer alignment: */
54#define MAX_SIZEOF_BTS (3 * 8) /* maximal size of a BTS record */ 63#define DS_ALIGNMENT (1 << 3)
55#define DS_ALIGNMENT (1 << 3) /* BTS and PEBS buffer alignment */
56 64
57#define BTS_CONTROL \ 65/* Number of buffer pointers in DS: */
58 (ds_cfg.ctl[dsf_bts] | ds_cfg.ctl[dsf_bts_kernel] | ds_cfg.ctl[dsf_bts_user] |\ 66#define NUM_DS_PTR_FIELDS 8
59 ds_cfg.ctl[dsf_bts_overflow])
60 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] )
61 77
62/* 78/*
63 * A BTS or PEBS tracer. 79 * A BTS or PEBS tracer.
@@ -66,29 +82,36 @@ static DEFINE_PER_CPU(struct ds_configuration, ds_cfg_array);
66 * to identify tracers. 82 * to identify tracers.
67 */ 83 */
68struct ds_tracer { 84struct ds_tracer {
69 /* the DS context (partially) owned by this tracer */ 85 /* The DS context (partially) owned by this tracer. */
70 struct ds_context *context; 86 struct ds_context *context;
71 /* the buffer provided on ds_request() and its size in bytes */ 87 /* The buffer provided on ds_request() and its size in bytes. */
72 void *buffer; 88 void *buffer;
73 size_t size; 89 size_t size;
74}; 90};
75 91
76struct bts_tracer { 92struct bts_tracer {
77 /* the common DS part */ 93 /* The common DS part: */
78 struct ds_tracer ds; 94 struct ds_tracer ds;
79 /* the trace including the DS configuration */ 95
80 struct bts_trace trace; 96 /* The trace including the DS configuration: */
81 /* buffer overflow notification function */ 97 struct bts_trace trace;
82 bts_ovfl_callback_t ovfl; 98
99 /* Buffer overflow notification function: */
100 bts_ovfl_callback_t ovfl;
101
102 /* Active flags affecting trace collection. */
103 unsigned int flags;
83}; 104};
84 105
85struct pebs_tracer { 106struct pebs_tracer {
86 /* the common DS part */ 107 /* The common DS part: */
87 struct ds_tracer ds; 108 struct ds_tracer ds;
88 /* the trace including the DS configuration */ 109
89 struct pebs_trace trace; 110 /* The trace including the DS configuration: */
90 /* buffer overflow notification function */ 111 struct pebs_trace trace;
91 pebs_ovfl_callback_t ovfl; 112
113 /* Buffer overflow notification function: */
114 pebs_ovfl_callback_t ovfl;
92}; 115};
93 116
94/* 117/*
@@ -97,6 +120,7 @@ struct pebs_tracer {
97 * 120 *
98 * The DS configuration consists of the following fields; different 121 * The DS configuration consists of the following fields; different
99 * architetures vary in the size of those fields. 122 * architetures vary in the size of those fields.
123 *
100 * - double-word aligned base linear address of the BTS buffer 124 * - double-word aligned base linear address of the BTS buffer
101 * - write pointer into the BTS buffer 125 * - write pointer into the BTS buffer
102 * - end linear address of the BTS buffer (one byte beyond the end of 126 * - end linear address of the BTS buffer (one byte beyond the end of
@@ -135,21 +159,22 @@ enum ds_field {
135}; 159};
136 160
137enum ds_qualifier { 161enum ds_qualifier {
138 ds_bts = 0, 162 ds_bts = 0,
139 ds_pebs 163 ds_pebs
140}; 164};
141 165
142static inline unsigned long ds_get(const unsigned char *base, 166static inline unsigned long
143 enum ds_qualifier qual, enum ds_field field) 167ds_get(const unsigned char *base, enum ds_qualifier qual, enum ds_field field)
144{ 168{
145 base += (ds_cfg.sizeof_field * (field + (4 * qual))); 169 base += (ds_cfg.sizeof_ptr_field * (field + (4 * qual)));
146 return *(unsigned long *)base; 170 return *(unsigned long *)base;
147} 171}
148 172
149static inline void ds_set(unsigned char *base, enum ds_qualifier qual, 173static inline void
150 enum ds_field field, unsigned long value) 174ds_set(unsigned char *base, enum ds_qualifier qual, enum ds_field field,
175 unsigned long value)
151{ 176{
152 base += (ds_cfg.sizeof_field * (field + (4 * qual))); 177 base += (ds_cfg.sizeof_ptr_field * (field + (4 * qual)));
153 (*(unsigned long *)base) = value; 178 (*(unsigned long *)base) = value;
154} 179}
155 180
@@ -159,7 +184,6 @@ static inline void ds_set(unsigned char *base, enum ds_qualifier qual,
159 */ 184 */
160static DEFINE_SPINLOCK(ds_lock); 185static DEFINE_SPINLOCK(ds_lock);
161 186
162
163/* 187/*
164 * We either support (system-wide) per-cpu or per-thread allocation. 188 * We either support (system-wide) per-cpu or per-thread allocation.
165 * We distinguish the two based on the task_struct pointer, where a 189 * We distinguish the two based on the task_struct pointer, where a
@@ -178,12 +202,28 @@ static DEFINE_SPINLOCK(ds_lock);
178 */ 202 */
179static atomic_t tracers = ATOMIC_INIT(0); 203static atomic_t tracers = ATOMIC_INIT(0);
180 204
181static inline void get_tracer(struct task_struct *task) 205static inline int get_tracer(struct task_struct *task)
182{ 206{
183 if (task) 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;
184 atomic_inc(&tracers); 215 atomic_inc(&tracers);
185 else 216 } else {
217 error = -EPERM;
218 if (atomic_read(&tracers) > 0)
219 goto out;
186 atomic_dec(&tracers); 220 atomic_dec(&tracers);
221 }
222
223 error = 0;
224out:
225 spin_unlock_irq(&ds_lock);
226 return error;
187} 227}
188 228
189static inline void put_tracer(struct task_struct *task) 229static inline void put_tracer(struct task_struct *task)
@@ -194,14 +234,6 @@ static inline void put_tracer(struct task_struct *task)
194 atomic_inc(&tracers); 234 atomic_inc(&tracers);
195} 235}
196 236
197static inline int check_tracer(struct task_struct *task)
198{
199 return task ?
200 (atomic_read(&tracers) >= 0) :
201 (atomic_read(&tracers) <= 0);
202}
203
204
205/* 237/*
206 * The DS context is either attached to a thread or to a cpu: 238 * The DS context is either attached to a thread or to a cpu:
207 * - in the former case, the thread_struct contains a pointer to the 239 * - in the former case, the thread_struct contains a pointer to the
@@ -213,61 +245,58 @@ static inline int check_tracer(struct task_struct *task)
213 * deallocated when the last user puts the context. 245 * deallocated when the last user puts the context.
214 */ 246 */
215struct ds_context { 247struct ds_context {
216 /* pointer to the DS configuration; goes into MSR_IA32_DS_AREA */ 248 /* The DS configuration; goes into MSR_IA32_DS_AREA: */
217 unsigned char ds[MAX_SIZEOF_DS]; 249 unsigned char ds[MAX_SIZEOF_DS];
218 /* the owner of the BTS and PEBS configuration, respectively */ 250
219 struct bts_tracer *bts_master; 251 /* The owner of the BTS and PEBS configuration, respectively: */
220 struct pebs_tracer *pebs_master; 252 struct bts_tracer *bts_master;
221 /* use count */ 253 struct pebs_tracer *pebs_master;
222 unsigned long count;
223 /* a pointer to the context location inside the thread_struct
224 * or the per_cpu context array */
225 struct ds_context **this;
226 /* a pointer to the task owning this context, or NULL, if the
227 * context is owned by a cpu */
228 struct task_struct *task;
229};
230 254
231static DEFINE_PER_CPU(struct ds_context *, system_context_array); 255 /* Use count: */
256 unsigned long count;
232 257
233#define system_context per_cpu(system_context_array, smp_processor_id()) 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};
234 267
268static DEFINE_PER_CPU(struct ds_context *, cpu_context);
235 269
236static inline struct ds_context *ds_get_context(struct task_struct *task) 270
271static struct ds_context *ds_get_context(struct task_struct *task, int cpu)
237{ 272{
238 struct ds_context **p_context = 273 struct ds_context **p_context =
239 (task ? &task->thread.ds_ctx : &system_context); 274 (task ? &task->thread.ds_ctx : &per_cpu(cpu_context, cpu));
240 struct ds_context *context = NULL; 275 struct ds_context *context = NULL;
241 struct ds_context *new_context = NULL; 276 struct ds_context *new_context = NULL;
242 unsigned long irq;
243 277
244 /* Chances are small that we already have a context. */ 278 /* Chances are small that we already have a context. */
245 new_context = kzalloc(sizeof(*new_context), GFP_KERNEL); 279 new_context = kzalloc(sizeof(*new_context), GFP_KERNEL);
246 if (!new_context) 280 if (!new_context)
247 return NULL; 281 return NULL;
248 282
249 spin_lock_irqsave(&ds_lock, irq); 283 spin_lock_irq(&ds_lock);
250 284
251 context = *p_context; 285 context = *p_context;
252 if (!context) { 286 if (likely(!context)) {
253 context = new_context; 287 context = new_context;
254 288
255 context->this = p_context; 289 context->this = p_context;
256 context->task = task; 290 context->task = task;
291 context->cpu = cpu;
257 context->count = 0; 292 context->count = 0;
258 293
259 if (task)
260 set_tsk_thread_flag(task, TIF_DS_AREA_MSR);
261
262 if (!task || (task == current))
263 wrmsrl(MSR_IA32_DS_AREA, (unsigned long)context->ds);
264
265 *p_context = context; 294 *p_context = context;
266 } 295 }
267 296
268 context->count++; 297 context->count++;
269 298
270 spin_unlock_irqrestore(&ds_lock, irq); 299 spin_unlock_irq(&ds_lock);
271 300
272 if (context != new_context) 301 if (context != new_context)
273 kfree(new_context); 302 kfree(new_context);
@@ -275,8 +304,9 @@ static inline struct ds_context *ds_get_context(struct task_struct *task)
275 return context; 304 return context;
276} 305}
277 306
278static inline void ds_put_context(struct ds_context *context) 307static void ds_put_context(struct ds_context *context)
279{ 308{
309 struct task_struct *task;
280 unsigned long irq; 310 unsigned long irq;
281 311
282 if (!context) 312 if (!context)
@@ -291,17 +321,55 @@ static inline void ds_put_context(struct ds_context *context)
291 321
292 *(context->this) = NULL; 322 *(context->this) = NULL;
293 323
294 if (context->task) 324 task = context->task;
295 clear_tsk_thread_flag(context->task, TIF_DS_AREA_MSR); 325
326 if (task)
327 clear_tsk_thread_flag(task, TIF_DS_AREA_MSR);
296 328
297 if (!context->task || (context->task == current)) 329 /*
298 wrmsrl(MSR_IA32_DS_AREA, 0); 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 */
299 338
300 spin_unlock_irqrestore(&ds_lock, irq); 339 spin_unlock_irqrestore(&ds_lock, irq);
301 340
341 /* The context might still be in use for context switching. */
342 if (task && (task != current))
343 wait_task_context_switch(task);
344
302 kfree(context); 345 kfree(context);
303} 346}
304 347
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}
305 373
306/* 374/*
307 * Call the tracer's callback on a buffer overflow. 375 * Call the tracer's callback on a buffer overflow.
@@ -332,9 +400,9 @@ static void ds_overflow(struct ds_context *context, enum ds_qualifier qual)
332 * The remainder of any partially written record is zeroed out. 400 * The remainder of any partially written record is zeroed out.
333 * 401 *
334 * context: the DS context 402 * context: the DS context
335 * qual: the buffer type 403 * qual: the buffer type
336 * record: the data to write 404 * record: the data to write
337 * size: the size of the data 405 * size: the size of the data
338 */ 406 */
339static int ds_write(struct ds_context *context, enum ds_qualifier qual, 407static int ds_write(struct ds_context *context, enum ds_qualifier qual,
340 const void *record, size_t size) 408 const void *record, size_t size)
@@ -349,14 +417,14 @@ static int ds_write(struct ds_context *context, enum ds_qualifier qual,
349 unsigned long write_size, adj_write_size; 417 unsigned long write_size, adj_write_size;
350 418
351 /* 419 /*
352 * write as much as possible without producing an 420 * Write as much as possible without producing an
353 * overflow interrupt. 421 * overflow interrupt.
354 * 422 *
355 * interrupt_threshold must either be 423 * Interrupt_threshold must either be
356 * - bigger than absolute_maximum or 424 * - bigger than absolute_maximum or
357 * - point to a record between buffer_base and absolute_maximum 425 * - point to a record between buffer_base and absolute_maximum
358 * 426 *
359 * index points to a valid record. 427 * Index points to a valid record.
360 */ 428 */
361 base = ds_get(context->ds, qual, ds_buffer_base); 429 base = ds_get(context->ds, qual, ds_buffer_base);
362 index = ds_get(context->ds, qual, ds_index); 430 index = ds_get(context->ds, qual, ds_index);
@@ -365,8 +433,10 @@ static int ds_write(struct ds_context *context, enum ds_qualifier qual,
365 433
366 write_end = min(end, int_th); 434 write_end = min(end, int_th);
367 435
368 /* if we are already beyond the interrupt threshold, 436 /*
369 * we fill the entire buffer */ 437 * If we are already beyond the interrupt threshold,
438 * we fill the entire buffer.
439 */
370 if (write_end <= index) 440 if (write_end <= index)
371 write_end = end; 441 write_end = end;
372 442
@@ -383,7 +453,7 @@ static int ds_write(struct ds_context *context, enum ds_qualifier qual,
383 adj_write_size = write_size / ds_cfg.sizeof_rec[qual]; 453 adj_write_size = write_size / ds_cfg.sizeof_rec[qual];
384 adj_write_size *= ds_cfg.sizeof_rec[qual]; 454 adj_write_size *= ds_cfg.sizeof_rec[qual];
385 455
386 /* zero out trailing bytes */ 456 /* Zero out trailing bytes. */
387 memset((char *)index + write_size, 0, 457 memset((char *)index + write_size, 0,
388 adj_write_size - write_size); 458 adj_write_size - write_size);
389 index += adj_write_size; 459 index += adj_write_size;
@@ -410,7 +480,7 @@ static int ds_write(struct ds_context *context, enum ds_qualifier qual,
410 * Later architectures use 64bit pointers throughout, whereas earlier 480 * Later architectures use 64bit pointers throughout, whereas earlier
411 * architectures use 32bit pointers in 32bit mode. 481 * architectures use 32bit pointers in 32bit mode.
412 * 482 *
413 * We compute the base address for the first 8 fields based on: 483 * We compute the base address for the fields based on:
414 * - the field size stored in the DS configuration 484 * - the field size stored in the DS configuration
415 * - the relative field position 485 * - the relative field position
416 * 486 *
@@ -431,23 +501,23 @@ enum bts_field {
431 bts_to, 501 bts_to,
432 bts_flags, 502 bts_flags,
433 503
434 bts_qual = bts_from, 504 bts_qual = bts_from,
435 bts_jiffies = bts_to, 505 bts_clock = bts_to,
436 bts_pid = bts_flags, 506 bts_pid = bts_flags,
437 507
438 bts_qual_mask = (bts_qual_max - 1), 508 bts_qual_mask = (bts_qual_max - 1),
439 bts_escape = ((unsigned long)-1 & ~bts_qual_mask) 509 bts_escape = ((unsigned long)-1 & ~bts_qual_mask)
440}; 510};
441 511
442static inline unsigned long bts_get(const char *base, enum bts_field field) 512static inline unsigned long bts_get(const char *base, enum bts_field field)
443{ 513{
444 base += (ds_cfg.sizeof_field * field); 514 base += (ds_cfg.sizeof_ptr_field * field);
445 return *(unsigned long *)base; 515 return *(unsigned long *)base;
446} 516}
447 517
448static inline void bts_set(char *base, enum bts_field field, unsigned long val) 518static inline void bts_set(char *base, enum bts_field field, unsigned long val)
449{ 519{
450 base += (ds_cfg.sizeof_field * field);; 520 base += (ds_cfg.sizeof_ptr_field * field);;
451 (*(unsigned long *)base) = val; 521 (*(unsigned long *)base) = val;
452} 522}
453 523
@@ -463,8 +533,8 @@ static inline void bts_set(char *base, enum bts_field field, unsigned long val)
463 * 533 *
464 * return: bytes read/written on success; -Eerrno, otherwise 534 * return: bytes read/written on success; -Eerrno, otherwise
465 */ 535 */
466static int bts_read(struct bts_tracer *tracer, const void *at, 536static int
467 struct bts_struct *out) 537bts_read(struct bts_tracer *tracer, const void *at, struct bts_struct *out)
468{ 538{
469 if (!tracer) 539 if (!tracer)
470 return -EINVAL; 540 return -EINVAL;
@@ -478,8 +548,8 @@ static int bts_read(struct bts_tracer *tracer, const void *at,
478 memset(out, 0, sizeof(*out)); 548 memset(out, 0, sizeof(*out));
479 if ((bts_get(at, bts_qual) & ~bts_qual_mask) == bts_escape) { 549 if ((bts_get(at, bts_qual) & ~bts_qual_mask) == bts_escape) {
480 out->qualifier = (bts_get(at, bts_qual) & bts_qual_mask); 550 out->qualifier = (bts_get(at, bts_qual) & bts_qual_mask);
481 out->variant.timestamp.jiffies = bts_get(at, bts_jiffies); 551 out->variant.event.clock = bts_get(at, bts_clock);
482 out->variant.timestamp.pid = bts_get(at, bts_pid); 552 out->variant.event.pid = bts_get(at, bts_pid);
483 } else { 553 } else {
484 out->qualifier = bts_branch; 554 out->qualifier = bts_branch;
485 out->variant.lbr.from = bts_get(at, bts_from); 555 out->variant.lbr.from = bts_get(at, bts_from);
@@ -516,8 +586,8 @@ static int bts_write(struct bts_tracer *tracer, const struct bts_struct *in)
516 case bts_task_arrives: 586 case bts_task_arrives:
517 case bts_task_departs: 587 case bts_task_departs:
518 bts_set(raw, bts_qual, (bts_escape | in->qualifier)); 588 bts_set(raw, bts_qual, (bts_escape | in->qualifier));
519 bts_set(raw, bts_jiffies, in->variant.timestamp.jiffies); 589 bts_set(raw, bts_clock, in->variant.event.clock);
520 bts_set(raw, bts_pid, in->variant.timestamp.pid); 590 bts_set(raw, bts_pid, in->variant.event.pid);
521 break; 591 break;
522 default: 592 default:
523 return -EINVAL; 593 return -EINVAL;
@@ -555,7 +625,8 @@ static void ds_init_ds_trace(struct ds_trace *trace, enum ds_qualifier qual,
555 unsigned int flags) { 625 unsigned int flags) {
556 unsigned long buffer, adj; 626 unsigned long buffer, adj;
557 627
558 /* adjust the buffer address and size to meet alignment 628 /*
629 * Adjust the buffer address and size to meet alignment
559 * constraints: 630 * constraints:
560 * - buffer is double-word aligned 631 * - buffer is double-word aligned
561 * - size is multiple of record size 632 * - size is multiple of record size
@@ -577,9 +648,11 @@ static void ds_init_ds_trace(struct ds_trace *trace, enum ds_qualifier qual,
577 trace->begin = (void *)buffer; 648 trace->begin = (void *)buffer;
578 trace->top = trace->begin; 649 trace->top = trace->begin;
579 trace->end = (void *)(buffer + size); 650 trace->end = (void *)(buffer + size);
580 /* The value for 'no threshold' is -1, which will set the 651 /*
652 * The value for 'no threshold' is -1, which will set the
581 * threshold outside of the buffer, just like we want it. 653 * threshold outside of the buffer, just like we want it.
582 */ 654 */
655 ith *= ds_cfg.sizeof_rec[qual];
583 trace->ith = (void *)(buffer + size - ith); 656 trace->ith = (void *)(buffer + size - ith);
584 657
585 trace->flags = flags; 658 trace->flags = flags;
@@ -588,18 +661,27 @@ static void ds_init_ds_trace(struct ds_trace *trace, enum ds_qualifier qual,
588 661
589static int ds_request(struct ds_tracer *tracer, struct ds_trace *trace, 662static int ds_request(struct ds_tracer *tracer, struct ds_trace *trace,
590 enum ds_qualifier qual, struct task_struct *task, 663 enum ds_qualifier qual, struct task_struct *task,
591 void *base, size_t size, size_t th, unsigned int flags) 664 int cpu, void *base, size_t size, size_t th)
592{ 665{
593 struct ds_context *context; 666 struct ds_context *context;
594 int error; 667 int error;
668 size_t req_size;
669
670 error = -EOPNOTSUPP;
671 if (!ds_cfg.sizeof_rec[qual])
672 goto out;
595 673
596 error = -EINVAL; 674 error = -EINVAL;
597 if (!base) 675 if (!base)
598 goto out; 676 goto out;
599 677
600 /* we require some space to do alignment adjustments below */ 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
601 error = -EINVAL; 683 error = -EINVAL;
602 if (size < (DS_ALIGNMENT + ds_cfg.sizeof_rec[qual])) 684 if (size < req_size)
603 goto out; 685 goto out;
604 686
605 if (th != (size_t)-1) { 687 if (th != (size_t)-1) {
@@ -614,182 +696,318 @@ static int ds_request(struct ds_tracer *tracer, struct ds_trace *trace,
614 tracer->size = size; 696 tracer->size = size;
615 697
616 error = -ENOMEM; 698 error = -ENOMEM;
617 context = ds_get_context(task); 699 context = ds_get_context(task, cpu);
618 if (!context) 700 if (!context)
619 goto out; 701 goto out;
620 tracer->context = context; 702 tracer->context = context;
621 703
622 ds_init_ds_trace(trace, qual, base, size, th, flags); 704 /*
705 * Defer any tracer-specific initialization work for the context until
706 * context ownership has been clarified.
707 */
623 708
624 error = 0; 709 error = 0;
625 out: 710 out:
626 return error; 711 return error;
627} 712}
628 713
629struct bts_tracer *ds_request_bts(struct task_struct *task, 714static struct bts_tracer *ds_request_bts(struct task_struct *task, int cpu,
630 void *base, size_t size, 715 void *base, size_t size,
631 bts_ovfl_callback_t ovfl, size_t th, 716 bts_ovfl_callback_t ovfl, size_t th,
632 unsigned int flags) 717 unsigned int flags)
633{ 718{
634 struct bts_tracer *tracer; 719 struct bts_tracer *tracer;
635 unsigned long irq;
636 int error; 720 int error;
637 721
722 /* Buffer overflow notification is not yet implemented. */
638 error = -EOPNOTSUPP; 723 error = -EOPNOTSUPP;
639 if (!ds_cfg.ctl[dsf_bts]) 724 if (ovfl)
640 goto out; 725 goto out;
641 726
642 /* buffer overflow notification is not yet implemented */ 727 error = get_tracer(task);
643 error = -EOPNOTSUPP; 728 if (error < 0)
644 if (ovfl)
645 goto out; 729 goto out;
646 730
647 error = -ENOMEM; 731 error = -ENOMEM;
648 tracer = kzalloc(sizeof(*tracer), GFP_KERNEL); 732 tracer = kzalloc(sizeof(*tracer), GFP_KERNEL);
649 if (!tracer) 733 if (!tracer)
650 goto out; 734 goto out_put_tracer;
651 tracer->ovfl = ovfl; 735 tracer->ovfl = ovfl;
652 736
737 /* Do some more error checking and acquire a tracing context. */
653 error = ds_request(&tracer->ds, &tracer->trace.ds, 738 error = ds_request(&tracer->ds, &tracer->trace.ds,
654 ds_bts, task, base, size, th, flags); 739 ds_bts, task, cpu, base, size, th);
655 if (error < 0) 740 if (error < 0)
656 goto out_tracer; 741 goto out_tracer;
657 742
658 743 /* Claim the bts part of the tracing context we acquired above. */
659 spin_lock_irqsave(&ds_lock, irq); 744 spin_lock_irq(&ds_lock);
660
661 error = -EPERM;
662 if (!check_tracer(task))
663 goto out_unlock;
664 get_tracer(task);
665 745
666 error = -EPERM; 746 error = -EPERM;
667 if (tracer->ds.context->bts_master) 747 if (tracer->ds.context->bts_master)
668 goto out_put_tracer; 748 goto out_unlock;
669 tracer->ds.context->bts_master = tracer; 749 tracer->ds.context->bts_master = tracer;
670 750
671 spin_unlock_irqrestore(&ds_lock, irq); 751 spin_unlock_irq(&ds_lock);
672 752
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);
673 760
674 tracer->trace.read = bts_read; 761 tracer->trace.read = bts_read;
675 tracer->trace.write = bts_write; 762 tracer->trace.write = bts_write;
676 763
677 ds_write_config(tracer->ds.context, &tracer->trace.ds, ds_bts); 764 /* Start tracing. */
678 ds_resume_bts(tracer); 765 ds_resume_bts(tracer);
679 766
680 return tracer; 767 return tracer;
681 768
682 out_put_tracer:
683 put_tracer(task);
684 out_unlock: 769 out_unlock:
685 spin_unlock_irqrestore(&ds_lock, irq); 770 spin_unlock_irq(&ds_lock);
686 ds_put_context(tracer->ds.context); 771 ds_put_context(tracer->ds.context);
687 out_tracer: 772 out_tracer:
688 kfree(tracer); 773 kfree(tracer);
774 out_put_tracer:
775 put_tracer(task);
689 out: 776 out:
690 return ERR_PTR(error); 777 return ERR_PTR(error);
691} 778}
692 779
693struct pebs_tracer *ds_request_pebs(struct task_struct *task, 780struct bts_tracer *ds_request_bts_task(struct task_struct *task,
694 void *base, size_t size, 781 void *base, size_t size,
695 pebs_ovfl_callback_t ovfl, size_t th, 782 bts_ovfl_callback_t ovfl,
696 unsigned int flags) 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)
697{ 799{
698 struct pebs_tracer *tracer; 800 struct pebs_tracer *tracer;
699 unsigned long irq;
700 int error; 801 int error;
701 802
702 /* buffer overflow notification is not yet implemented */ 803 /* Buffer overflow notification is not yet implemented. */
703 error = -EOPNOTSUPP; 804 error = -EOPNOTSUPP;
704 if (ovfl) 805 if (ovfl)
705 goto out; 806 goto out;
706 807
808 error = get_tracer(task);
809 if (error < 0)
810 goto out;
811
707 error = -ENOMEM; 812 error = -ENOMEM;
708 tracer = kzalloc(sizeof(*tracer), GFP_KERNEL); 813 tracer = kzalloc(sizeof(*tracer), GFP_KERNEL);
709 if (!tracer) 814 if (!tracer)
710 goto out; 815 goto out_put_tracer;
711 tracer->ovfl = ovfl; 816 tracer->ovfl = ovfl;
712 817
818 /* Do some more error checking and acquire a tracing context. */
713 error = ds_request(&tracer->ds, &tracer->trace.ds, 819 error = ds_request(&tracer->ds, &tracer->trace.ds,
714 ds_pebs, task, base, size, th, flags); 820 ds_pebs, task, cpu, base, size, th);
715 if (error < 0) 821 if (error < 0)
716 goto out_tracer; 822 goto out_tracer;
717 823
718 spin_lock_irqsave(&ds_lock, irq); 824 /* Claim the pebs part of the tracing context we acquired above. */
719 825 spin_lock_irq(&ds_lock);
720 error = -EPERM;
721 if (!check_tracer(task))
722 goto out_unlock;
723 get_tracer(task);
724 826
725 error = -EPERM; 827 error = -EPERM;
726 if (tracer->ds.context->pebs_master) 828 if (tracer->ds.context->pebs_master)
727 goto out_put_tracer; 829 goto out_unlock;
728 tracer->ds.context->pebs_master = tracer; 830 tracer->ds.context->pebs_master = tracer;
729 831
730 spin_unlock_irqrestore(&ds_lock, irq); 832 spin_unlock_irq(&ds_lock);
731 833
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);
732 ds_write_config(tracer->ds.context, &tracer->trace.ds, ds_pebs); 839 ds_write_config(tracer->ds.context, &tracer->trace.ds, ds_pebs);
840 ds_install_ds_area(tracer->ds.context);
841
842 /* Start tracing. */
733 ds_resume_pebs(tracer); 843 ds_resume_pebs(tracer);
734 844
735 return tracer; 845 return tracer;
736 846
737 out_put_tracer:
738 put_tracer(task);
739 out_unlock: 847 out_unlock:
740 spin_unlock_irqrestore(&ds_lock, irq); 848 spin_unlock_irq(&ds_lock);
741 ds_put_context(tracer->ds.context); 849 ds_put_context(tracer->ds.context);
742 out_tracer: 850 out_tracer:
743 kfree(tracer); 851 kfree(tracer);
852 out_put_tracer:
853 put_tracer(task);
744 out: 854 out:
745 return ERR_PTR(error); 855 return ERR_PTR(error);
746} 856}
747 857
748void ds_release_bts(struct bts_tracer *tracer) 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)
749{ 862{
750 if (!tracer) 863 return ds_request_pebs(task, 0, base, size, ovfl, th, flags);
751 return; 864}
752 865
753 ds_suspend_bts(tracer); 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;
754 878
755 WARN_ON_ONCE(tracer->ds.context->bts_master != tracer); 879 WARN_ON_ONCE(tracer->ds.context->bts_master != tracer);
756 tracer->ds.context->bts_master = NULL; 880 tracer->ds.context->bts_master = NULL;
757 881
758 put_tracer(tracer->ds.context->task); 882 /* Make sure tracing stopped and the tracer is not in use. */
883 if (task && (task != current))
884 wait_task_context_switch(task);
885
759 ds_put_context(tracer->ds.context); 886 ds_put_context(tracer->ds.context);
887 put_tracer(task);
760 888
761 kfree(tracer); 889 kfree(tracer);
762} 890}
763 891
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
764void ds_suspend_bts(struct bts_tracer *tracer) 945void ds_suspend_bts(struct bts_tracer *tracer)
765{ 946{
766 struct task_struct *task; 947 struct task_struct *task;
948 unsigned long debugctlmsr;
949 int cpu;
767 950
768 if (!tracer) 951 if (!tracer)
769 return; 952 return;
770 953
954 tracer->flags = 0;
955
771 task = tracer->ds.context->task; 956 task = tracer->ds.context->task;
957 cpu = tracer->ds.context->cpu;
772 958
773 if (!task || (task == current)) 959 WARN_ON(!task && irqs_disabled());
774 update_debugctlmsr(get_debugctlmsr() & ~BTS_CONTROL);
775 960
776 if (task) { 961 debugctlmsr = (task ?
777 task->thread.debugctlmsr &= ~BTS_CONTROL; 962 task->thread.debugctlmsr :
963 get_debugctlmsr_on_cpu(cpu));
964 debugctlmsr &= ~BTS_CONTROL;
778 965
779 if (!task->thread.debugctlmsr) 966 if (task)
780 clear_tsk_thread_flag(task, TIF_DEBUGCTLMSR); 967 update_task_debugctlmsr(task, debugctlmsr);
781 } 968 else
969 update_debugctlmsr_on_cpu(cpu, debugctlmsr);
782} 970}
783 971
784void ds_resume_bts(struct bts_tracer *tracer) 972int ds_suspend_bts_noirq(struct bts_tracer *tracer)
785{ 973{
786 struct task_struct *task; 974 struct task_struct *task;
787 unsigned long control; 975 unsigned long debugctlmsr, irq;
976 int cpu, error = 0;
788 977
789 if (!tracer) 978 if (!tracer)
790 return; 979 return 0;
980
981 tracer->flags = 0;
791 982
792 task = tracer->ds.context->task; 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;
793 1011
794 control = ds_cfg.ctl[dsf_bts]; 1012 control = ds_cfg.ctl[dsf_bts];
795 if (!(tracer->trace.ds.flags & BTS_KERNEL)) 1013 if (!(tracer->trace.ds.flags & BTS_KERNEL))
@@ -797,41 +1015,149 @@ void ds_resume_bts(struct bts_tracer *tracer)
797 if (!(tracer->trace.ds.flags & BTS_USER)) 1015 if (!(tracer->trace.ds.flags & BTS_USER))
798 control |= ds_cfg.ctl[dsf_bts_user]; 1016 control |= ds_cfg.ctl[dsf_bts_user];
799 1017
800 if (task) { 1018 return control;
801 task->thread.debugctlmsr |= control;
802 set_tsk_thread_flag(task, TIF_DEBUGCTLMSR);
803 }
804
805 if (!task || (task == current))
806 update_debugctlmsr(get_debugctlmsr() | control);
807} 1019}
808 1020
809void ds_release_pebs(struct pebs_tracer *tracer) 1021void ds_resume_bts(struct bts_tracer *tracer)
810{ 1022{
1023 struct task_struct *task;
1024 unsigned long debugctlmsr;
1025 int cpu;
1026
811 if (!tracer) 1027 if (!tracer)
812 return; 1028 return;
813 1029
814 ds_suspend_pebs(tracer); 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;
815 1089
816 WARN_ON_ONCE(tracer->ds.context->pebs_master != tracer); 1090 WARN_ON_ONCE(tracer->ds.context->pebs_master != tracer);
817 tracer->ds.context->pebs_master = NULL; 1091 tracer->ds.context->pebs_master = NULL;
818 1092
819 put_tracer(tracer->ds.context->task);
820 ds_put_context(tracer->ds.context); 1093 ds_put_context(tracer->ds.context);
1094 put_tracer(task);
821 1095
822 kfree(tracer); 1096 kfree(tracer);
823} 1097}
824 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
825void ds_suspend_pebs(struct pebs_tracer *tracer) 1141void ds_suspend_pebs(struct pebs_tracer *tracer)
826{ 1142{
827 1143
828} 1144}
829 1145
1146int ds_suspend_pebs_noirq(struct pebs_tracer *tracer)
1147{
1148 return 0;
1149}
1150
830void ds_resume_pebs(struct pebs_tracer *tracer) 1151void ds_resume_pebs(struct pebs_tracer *tracer)
831{ 1152{
832 1153
833} 1154}
834 1155
1156int ds_resume_pebs_noirq(struct pebs_tracer *tracer)
1157{
1158 return 0;
1159}
1160
835const struct bts_trace *ds_read_bts(struct bts_tracer *tracer) 1161const struct bts_trace *ds_read_bts(struct bts_tracer *tracer)
836{ 1162{
837 if (!tracer) 1163 if (!tracer)
@@ -847,8 +1173,12 @@ const struct pebs_trace *ds_read_pebs(struct pebs_tracer *tracer)
847 return NULL; 1173 return NULL;
848 1174
849 ds_read_config(tracer->ds.context, &tracer->trace.ds, ds_pebs); 1175 ds_read_config(tracer->ds.context, &tracer->trace.ds, ds_pebs);
850 tracer->trace.reset_value = 1176
851 *(u64 *)(tracer->ds.context->ds + (ds_cfg.sizeof_field * 8)); 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);
852 1182
853 return &tracer->trace; 1183 return &tracer->trace;
854} 1184}
@@ -873,18 +1203,24 @@ int ds_reset_pebs(struct pebs_tracer *tracer)
873 1203
874 tracer->trace.ds.top = tracer->trace.ds.begin; 1204 tracer->trace.ds.top = tracer->trace.ds.begin;
875 1205
876 ds_set(tracer->ds.context->ds, ds_bts, ds_index, 1206 ds_set(tracer->ds.context->ds, ds_pebs, ds_index,
877 (unsigned long)tracer->trace.ds.top); 1207 (unsigned long)tracer->trace.ds.top);
878 1208
879 return 0; 1209 return 0;
880} 1210}
881 1211
882int ds_set_pebs_reset(struct pebs_tracer *tracer, u64 value) 1212int ds_set_pebs_reset(struct pebs_tracer *tracer,
1213 unsigned int counter, u64 value)
883{ 1214{
884 if (!tracer) 1215 if (!tracer)
885 return -EINVAL; 1216 return -EINVAL;
886 1217
887 *(u64 *)(tracer->ds.context->ds + (ds_cfg.sizeof_field * 8)) = value; 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;
888 1224
889 return 0; 1225 return 0;
890} 1226}
@@ -894,73 +1230,117 @@ static const struct ds_configuration ds_cfg_netburst = {
894 .ctl[dsf_bts] = (1 << 2) | (1 << 3), 1230 .ctl[dsf_bts] = (1 << 2) | (1 << 3),
895 .ctl[dsf_bts_kernel] = (1 << 5), 1231 .ctl[dsf_bts_kernel] = (1 << 5),
896 .ctl[dsf_bts_user] = (1 << 6), 1232 .ctl[dsf_bts_user] = (1 << 6),
897 1233 .nr_counter_reset = 1,
898 .sizeof_field = sizeof(long),
899 .sizeof_rec[ds_bts] = sizeof(long) * 3,
900#ifdef __i386__
901 .sizeof_rec[ds_pebs] = sizeof(long) * 10,
902#else
903 .sizeof_rec[ds_pebs] = sizeof(long) * 18,
904#endif
905}; 1234};
906static const struct ds_configuration ds_cfg_pentium_m = { 1235static const struct ds_configuration ds_cfg_pentium_m = {
907 .name = "Pentium M", 1236 .name = "Pentium M",
908 .ctl[dsf_bts] = (1 << 6) | (1 << 7), 1237 .ctl[dsf_bts] = (1 << 6) | (1 << 7),
909 1238 .nr_counter_reset = 1,
910 .sizeof_field = sizeof(long),
911 .sizeof_rec[ds_bts] = sizeof(long) * 3,
912#ifdef __i386__
913 .sizeof_rec[ds_pebs] = sizeof(long) * 10,
914#else
915 .sizeof_rec[ds_pebs] = sizeof(long) * 18,
916#endif
917}; 1239};
918static const struct ds_configuration ds_cfg_core2_atom = { 1240static const struct ds_configuration ds_cfg_core2_atom = {
919 .name = "Core 2/Atom", 1241 .name = "Core 2/Atom",
920 .ctl[dsf_bts] = (1 << 6) | (1 << 7), 1242 .ctl[dsf_bts] = (1 << 6) | (1 << 7),
921 .ctl[dsf_bts_kernel] = (1 << 9), 1243 .ctl[dsf_bts_kernel] = (1 << 9),
922 .ctl[dsf_bts_user] = (1 << 10), 1244 .ctl[dsf_bts_user] = (1 << 10),
923 1245 .nr_counter_reset = 1,
924 .sizeof_field = 8, 1246};
925 .sizeof_rec[ds_bts] = 8 * 3, 1247static const struct ds_configuration ds_cfg_core_i7 = {
926 .sizeof_rec[ds_pebs] = 8 * 18, 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,
927}; 1253};
928 1254
929static void 1255static void
930ds_configure(const struct ds_configuration *cfg) 1256ds_configure(const struct ds_configuration *cfg,
1257 struct cpuinfo_x86 *cpu)
931{ 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
932 memset(&ds_cfg, 0, sizeof(ds_cfg)); 1295 memset(&ds_cfg, 0, sizeof(ds_cfg));
933 ds_cfg = *cfg; 1296 ds_cfg = *cfg;
934 1297
935 printk(KERN_INFO "[ds] using %s configuration\n", ds_cfg.name); 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;
936 1303
937 if (!cpu_has_bts) { 1304 if (!cpu_has(cpu, X86_FEATURE_BTS)) {
938 ds_cfg.ctl[dsf_bts] = 0; 1305 ds_cfg.sizeof_rec[ds_bts] = 0;
939 printk(KERN_INFO "[ds] bts not available\n"); 1306 printk(KERN_INFO "[ds] bts not available\n");
940 } 1307 }
941 if (!cpu_has_pebs) 1308 if (!cpu_has(cpu, X86_FEATURE_PEBS)) {
1309 ds_cfg.sizeof_rec[ds_pebs] = 0;
942 printk(KERN_INFO "[ds] pebs not available\n"); 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]);
943 1317
944 WARN_ON_ONCE(MAX_SIZEOF_DS < (12 * ds_cfg.sizeof_field)); 1318 WARN_ON_ONCE(MAX_PEBS_COUNTERS < ds_cfg.nr_counter_reset);
945} 1319}
946 1320
947void __cpuinit ds_init_intel(struct cpuinfo_x86 *c) 1321void __cpuinit ds_init_intel(struct cpuinfo_x86 *c)
948{ 1322{
1323 /* Only configure the first cpu. Others are identical. */
1324 if (ds_cfg.name)
1325 return;
1326
949 switch (c->x86) { 1327 switch (c->x86) {
950 case 0x6: 1328 case 0x6:
951 switch (c->x86_model) { 1329 switch (c->x86_model) {
952 case 0x9: 1330 case 0x9:
953 case 0xd: /* Pentium M */ 1331 case 0xd: /* Pentium M */
954 ds_configure(&ds_cfg_pentium_m); 1332 ds_configure(&ds_cfg_pentium_m, c);
955 break; 1333 break;
956 case 0xf: 1334 case 0xf:
957 case 0x17: /* Core2 */ 1335 case 0x17: /* Core2 */
958 case 0x1c: /* Atom */ 1336 case 0x1c: /* Atom */
959 ds_configure(&ds_cfg_core2_atom); 1337 ds_configure(&ds_cfg_core2_atom, c);
1338 break;
1339 case 0x1a: /* Core i7 */
1340 ds_configure(&ds_cfg_core_i7, c);
960 break; 1341 break;
961 case 0x1a: /* i7 */
962 default: 1342 default:
963 /* sorry, don't know about them */ 1343 /* Sorry, don't know about them. */
964 break; 1344 break;
965 } 1345 }
966 break; 1346 break;
@@ -969,64 +1349,89 @@ void __cpuinit ds_init_intel(struct cpuinfo_x86 *c)
969 case 0x0: 1349 case 0x0:
970 case 0x1: 1350 case 0x1:
971 case 0x2: /* Netburst */ 1351 case 0x2: /* Netburst */
972 ds_configure(&ds_cfg_netburst); 1352 ds_configure(&ds_cfg_netburst, c);
973 break; 1353 break;
974 default: 1354 default:
975 /* sorry, don't know about them */ 1355 /* Sorry, don't know about them. */
976 break; 1356 break;
977 } 1357 }
978 break; 1358 break;
979 default: 1359 default:
980 /* sorry, don't know about them */ 1360 /* Sorry, don't know about them. */
981 break; 1361 break;
982 } 1362 }
983} 1363}
984 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
985/* 1386/*
986 * Change the DS configuration from tracing prev to tracing next. 1387 * Change the DS configuration from tracing prev to tracing next.
987 */ 1388 */
988void ds_switch_to(struct task_struct *prev, struct task_struct *next) 1389void ds_switch_to(struct task_struct *prev, struct task_struct *next)
989{ 1390{
990 struct ds_context *prev_ctx = prev->thread.ds_ctx; 1391 struct ds_context *prev_ctx = prev->thread.ds_ctx;
991 struct ds_context *next_ctx = next->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();
992 1397
993 if (prev_ctx) { 1398 if (prev_ctx) {
994 update_debugctlmsr(0); 1399 update_debugctlmsr(0);
995 1400
996 if (prev_ctx->bts_master && 1401 ds_take_timestamp(prev_ctx, bts_task_departs, prev);
997 (prev_ctx->bts_master->trace.ds.flags & BTS_TIMESTAMPS)) {
998 struct bts_struct ts = {
999 .qualifier = bts_task_departs,
1000 .variant.timestamp.jiffies = jiffies_64,
1001 .variant.timestamp.pid = prev->pid
1002 };
1003 bts_write(prev_ctx->bts_master, &ts);
1004 }
1005 } 1402 }
1006 1403
1007 if (next_ctx) { 1404 if (next_ctx) {
1008 if (next_ctx->bts_master && 1405 ds_take_timestamp(next_ctx, bts_task_arrives, next);
1009 (next_ctx->bts_master->trace.ds.flags & BTS_TIMESTAMPS)) {
1010 struct bts_struct ts = {
1011 .qualifier = bts_task_arrives,
1012 .variant.timestamp.jiffies = jiffies_64,
1013 .variant.timestamp.pid = next->pid
1014 };
1015 bts_write(next_ctx->bts_master, &ts);
1016 }
1017 1406
1018 wrmsrl(MSR_IA32_DS_AREA, (unsigned long)next_ctx->ds); 1407 wrmsrl(MSR_IA32_DS_AREA, (unsigned long)next_ctx->ds);
1019 } 1408 }
1020 1409
1021 update_debugctlmsr(next->thread.debugctlmsr); 1410 update_debugctlmsr(debugctlmsr);
1022} 1411}
1023 1412
1024void ds_copy_thread(struct task_struct *tsk, struct task_struct *father) 1413static __init int ds_selftest(void)
1025{ 1414{
1026 clear_tsk_thread_flag(tsk, TIF_DS_AREA_MSR); 1415 if (ds_cfg.sizeof_rec[ds_bts]) {
1027 tsk->thread.ds_ctx = NULL; 1416 int error;
1028}
1029 1417
1030void ds_exit_thread(struct task_struct *tsk) 1418 error = ds_selftest_bts();
1031{ 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;
1032} 1436}
1437device_initcall(ds_selftest);
diff --git a/arch/x86/kernel/ds_selftest.c b/arch/x86/kernel/ds_selftest.c
new file mode 100644
index 000000000000..6bc7c199ab99
--- /dev/null
+++ b/arch/x86/kernel/ds_selftest.c
@@ -0,0 +1,408 @@
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
new file mode 100644
index 000000000000..2ba8745c6663
--- /dev/null
+++ b/arch/x86/kernel/ds_selftest.h
@@ -0,0 +1,15 @@
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/entry_64.S b/arch/x86/kernel/entry_64.S
index bb01ce080b80..1c17d7c751a4 100644
--- a/arch/x86/kernel/entry_64.S
+++ b/arch/x86/kernel/entry_64.S
@@ -147,27 +147,14 @@ END(ftrace_graph_caller)
147GLOBAL(return_to_handler) 147GLOBAL(return_to_handler)
148 subq $80, %rsp 148 subq $80, %rsp
149 149
150 /* Save the return values */
150 movq %rax, (%rsp) 151 movq %rax, (%rsp)
151 movq %rcx, 8(%rsp) 152 movq %rdx, 8(%rsp)
152 movq %rdx, 16(%rsp)
153 movq %rsi, 24(%rsp)
154 movq %rdi, 32(%rsp)
155 movq %r8, 40(%rsp)
156 movq %r9, 48(%rsp)
157 movq %r10, 56(%rsp)
158 movq %r11, 64(%rsp)
159 153
160 call ftrace_return_to_handler 154 call ftrace_return_to_handler
161 155
162 movq %rax, 72(%rsp) 156 movq %rax, 72(%rsp)
163 movq 64(%rsp), %r11 157 movq 8(%rsp), %rdx
164 movq 56(%rsp), %r10
165 movq 48(%rsp), %r9
166 movq 40(%rsp), %r8
167 movq 32(%rsp), %rdi
168 movq 24(%rsp), %rsi
169 movq 16(%rsp), %rdx
170 movq 8(%rsp), %rcx
171 movq (%rsp), %rax 158 movq (%rsp), %rax
172 addq $72, %rsp 159 addq $72, %rsp
173 retq 160 retq
diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c
index e22d63bdc8ff..3bb2be1649bd 100644
--- a/arch/x86/kernel/process.c
+++ b/arch/x86/kernel/process.c
@@ -16,6 +16,7 @@
16#include <asm/idle.h> 16#include <asm/idle.h>
17#include <asm/uaccess.h> 17#include <asm/uaccess.h>
18#include <asm/i387.h> 18#include <asm/i387.h>
19#include <asm/ds.h>
19 20
20unsigned long idle_halt; 21unsigned long idle_halt;
21EXPORT_SYMBOL(idle_halt); 22EXPORT_SYMBOL(idle_halt);
@@ -47,6 +48,8 @@ void free_thread_xstate(struct task_struct *tsk)
47 kmem_cache_free(task_xstate_cachep, tsk->thread.xstate); 48 kmem_cache_free(task_xstate_cachep, tsk->thread.xstate);
48 tsk->thread.xstate = NULL; 49 tsk->thread.xstate = NULL;
49 } 50 }
51
52 WARN(tsk->thread.ds_ctx, "leaking DS context\n");
50} 53}
51 54
52void free_thread_info(struct thread_info *ti) 55void free_thread_info(struct thread_info *ti)
@@ -85,8 +88,6 @@ void exit_thread(void)
85 put_cpu(); 88 put_cpu();
86 kfree(bp); 89 kfree(bp);
87 } 90 }
88
89 ds_exit_thread(current);
90} 91}
91 92
92void flush_thread(void) 93void flush_thread(void)
diff --git a/arch/x86/kernel/process_32.c b/arch/x86/kernel/process_32.c
index c60924b5d123..59f4524984af 100644
--- a/arch/x86/kernel/process_32.c
+++ b/arch/x86/kernel/process_32.c
@@ -287,7 +287,8 @@ int copy_thread(unsigned long clone_flags, unsigned long sp,
287 p->thread.io_bitmap_max = 0; 287 p->thread.io_bitmap_max = 0;
288 } 288 }
289 289
290 ds_copy_thread(p, current); 290 clear_tsk_thread_flag(p, TIF_DS_AREA_MSR);
291 p->thread.ds_ctx = NULL;
291 292
292 clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR); 293 clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR);
293 p->thread.debugctlmsr = 0; 294 p->thread.debugctlmsr = 0;
diff --git a/arch/x86/kernel/process_64.c b/arch/x86/kernel/process_64.c
index 45f010fb2e20..ebefb5407b9d 100644
--- a/arch/x86/kernel/process_64.c
+++ b/arch/x86/kernel/process_64.c
@@ -332,7 +332,8 @@ int copy_thread(unsigned long clone_flags, unsigned long sp,
332 goto out; 332 goto out;
333 } 333 }
334 334
335 ds_copy_thread(p, me); 335 clear_tsk_thread_flag(p, TIF_DS_AREA_MSR);
336 p->thread.ds_ctx = NULL;
336 337
337 clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR); 338 clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR);
338 p->thread.debugctlmsr = 0; 339 p->thread.debugctlmsr = 0;
diff --git a/arch/x86/kernel/ptrace.c b/arch/x86/kernel/ptrace.c
index 23b7c8f017e2..09ecbde91c13 100644
--- a/arch/x86/kernel/ptrace.c
+++ b/arch/x86/kernel/ptrace.c
@@ -21,6 +21,7 @@
21#include <linux/audit.h> 21#include <linux/audit.h>
22#include <linux/seccomp.h> 22#include <linux/seccomp.h>
23#include <linux/signal.h> 23#include <linux/signal.h>
24#include <linux/workqueue.h>
24 25
25#include <asm/uaccess.h> 26#include <asm/uaccess.h>
26#include <asm/pgtable.h> 27#include <asm/pgtable.h>
@@ -578,17 +579,130 @@ static int ioperm_get(struct task_struct *target,
578} 579}
579 580
580#ifdef CONFIG_X86_PTRACE_BTS 581#ifdef CONFIG_X86_PTRACE_BTS
582/*
583 * A branch trace store context.
584 *
585 * Contexts may only be installed by ptrace_bts_config() and only for
586 * ptraced tasks.
587 *
588 * Contexts are destroyed when the tracee is detached from the tracer.
589 * The actual destruction work requires interrupts enabled, so the
590 * work is deferred and will be scheduled during __ptrace_unlink().
591 *
592 * Contexts hold an additional task_struct reference on the traced
593 * task, as well as a reference on the tracer's mm.
594 *
595 * Ptrace already holds a task_struct for the duration of ptrace operations,
596 * but since destruction is deferred, it may be executed after both
597 * tracer and tracee exited.
598 */
599struct bts_context {
600 /* The branch trace handle. */
601 struct bts_tracer *tracer;
602
603 /* The buffer used to store the branch trace and its size. */
604 void *buffer;
605 unsigned int size;
606
607 /* The mm that paid for the above buffer. */
608 struct mm_struct *mm;
609
610 /* The task this context belongs to. */
611 struct task_struct *task;
612
613 /* The signal to send on a bts buffer overflow. */
614 unsigned int bts_ovfl_signal;
615
616 /* The work struct to destroy a context. */
617 struct work_struct work;
618};
619
620static int alloc_bts_buffer(struct bts_context *context, unsigned int size)
621{
622 void *buffer = NULL;
623 int err = -ENOMEM;
624
625 err = account_locked_memory(current->mm, current->signal->rlim, size);
626 if (err < 0)
627 return err;
628
629 buffer = kzalloc(size, GFP_KERNEL);
630 if (!buffer)
631 goto out_refund;
632
633 context->buffer = buffer;
634 context->size = size;
635 context->mm = get_task_mm(current);
636
637 return 0;
638
639 out_refund:
640 refund_locked_memory(current->mm, size);
641 return err;
642}
643
644static inline void free_bts_buffer(struct bts_context *context)
645{
646 if (!context->buffer)
647 return;
648
649 kfree(context->buffer);
650 context->buffer = NULL;
651
652 refund_locked_memory(context->mm, context->size);
653 context->size = 0;
654
655 mmput(context->mm);
656 context->mm = NULL;
657}
658
659static void free_bts_context_work(struct work_struct *w)
660{
661 struct bts_context *context;
662
663 context = container_of(w, struct bts_context, work);
664
665 ds_release_bts(context->tracer);
666 put_task_struct(context->task);
667 free_bts_buffer(context);
668 kfree(context);
669}
670
671static inline void free_bts_context(struct bts_context *context)
672{
673 INIT_WORK(&context->work, free_bts_context_work);
674 schedule_work(&context->work);
675}
676
677static inline struct bts_context *alloc_bts_context(struct task_struct *task)
678{
679 struct bts_context *context = kzalloc(sizeof(*context), GFP_KERNEL);
680 if (context) {
681 context->task = task;
682 task->bts = context;
683
684 get_task_struct(task);
685 }
686
687 return context;
688}
689
581static int ptrace_bts_read_record(struct task_struct *child, size_t index, 690static int ptrace_bts_read_record(struct task_struct *child, size_t index,
582 struct bts_struct __user *out) 691 struct bts_struct __user *out)
583{ 692{
693 struct bts_context *context;
584 const struct bts_trace *trace; 694 const struct bts_trace *trace;
585 struct bts_struct bts; 695 struct bts_struct bts;
586 const unsigned char *at; 696 const unsigned char *at;
587 int error; 697 int error;
588 698
589 trace = ds_read_bts(child->bts); 699 context = child->bts;
700 if (!context)
701 return -ESRCH;
702
703 trace = ds_read_bts(context->tracer);
590 if (!trace) 704 if (!trace)
591 return -EPERM; 705 return -ESRCH;
592 706
593 at = trace->ds.top - ((index + 1) * trace->ds.size); 707 at = trace->ds.top - ((index + 1) * trace->ds.size);
594 if ((void *)at < trace->ds.begin) 708 if ((void *)at < trace->ds.begin)
@@ -597,7 +711,7 @@ static int ptrace_bts_read_record(struct task_struct *child, size_t index,
597 if (!trace->read) 711 if (!trace->read)
598 return -EOPNOTSUPP; 712 return -EOPNOTSUPP;
599 713
600 error = trace->read(child->bts, at, &bts); 714 error = trace->read(context->tracer, at, &bts);
601 if (error < 0) 715 if (error < 0)
602 return error; 716 return error;
603 717
@@ -611,13 +725,18 @@ static int ptrace_bts_drain(struct task_struct *child,
611 long size, 725 long size,
612 struct bts_struct __user *out) 726 struct bts_struct __user *out)
613{ 727{
728 struct bts_context *context;
614 const struct bts_trace *trace; 729 const struct bts_trace *trace;
615 const unsigned char *at; 730 const unsigned char *at;
616 int error, drained = 0; 731 int error, drained = 0;
617 732
618 trace = ds_read_bts(child->bts); 733 context = child->bts;
734 if (!context)
735 return -ESRCH;
736
737 trace = ds_read_bts(context->tracer);
619 if (!trace) 738 if (!trace)
620 return -EPERM; 739 return -ESRCH;
621 740
622 if (!trace->read) 741 if (!trace->read)
623 return -EOPNOTSUPP; 742 return -EOPNOTSUPP;
@@ -628,9 +747,8 @@ static int ptrace_bts_drain(struct task_struct *child,
628 for (at = trace->ds.begin; (void *)at < trace->ds.top; 747 for (at = trace->ds.begin; (void *)at < trace->ds.top;
629 out++, drained++, at += trace->ds.size) { 748 out++, drained++, at += trace->ds.size) {
630 struct bts_struct bts; 749 struct bts_struct bts;
631 int error;
632 750
633 error = trace->read(child->bts, at, &bts); 751 error = trace->read(context->tracer, at, &bts);
634 if (error < 0) 752 if (error < 0)
635 return error; 753 return error;
636 754
@@ -640,35 +758,18 @@ static int ptrace_bts_drain(struct task_struct *child,
640 758
641 memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size); 759 memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
642 760
643 error = ds_reset_bts(child->bts); 761 error = ds_reset_bts(context->tracer);
644 if (error < 0) 762 if (error < 0)
645 return error; 763 return error;
646 764
647 return drained; 765 return drained;
648} 766}
649 767
650static int ptrace_bts_allocate_buffer(struct task_struct *child, size_t size)
651{
652 child->bts_buffer = alloc_locked_buffer(size);
653 if (!child->bts_buffer)
654 return -ENOMEM;
655
656 child->bts_size = size;
657
658 return 0;
659}
660
661static void ptrace_bts_free_buffer(struct task_struct *child)
662{
663 free_locked_buffer(child->bts_buffer, child->bts_size);
664 child->bts_buffer = NULL;
665 child->bts_size = 0;
666}
667
668static int ptrace_bts_config(struct task_struct *child, 768static int ptrace_bts_config(struct task_struct *child,
669 long cfg_size, 769 long cfg_size,
670 const struct ptrace_bts_config __user *ucfg) 770 const struct ptrace_bts_config __user *ucfg)
671{ 771{
772 struct bts_context *context;
672 struct ptrace_bts_config cfg; 773 struct ptrace_bts_config cfg;
673 unsigned int flags = 0; 774 unsigned int flags = 0;
674 775
@@ -678,28 +779,33 @@ static int ptrace_bts_config(struct task_struct *child,
678 if (copy_from_user(&cfg, ucfg, sizeof(cfg))) 779 if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
679 return -EFAULT; 780 return -EFAULT;
680 781
681 if (child->bts) { 782 context = child->bts;
682 ds_release_bts(child->bts); 783 if (!context)
683 child->bts = NULL; 784 context = alloc_bts_context(child);
684 } 785 if (!context)
786 return -ENOMEM;
685 787
686 if (cfg.flags & PTRACE_BTS_O_SIGNAL) { 788 if (cfg.flags & PTRACE_BTS_O_SIGNAL) {
687 if (!cfg.signal) 789 if (!cfg.signal)
688 return -EINVAL; 790 return -EINVAL;
689 791
690 child->thread.bts_ovfl_signal = cfg.signal;
691 return -EOPNOTSUPP; 792 return -EOPNOTSUPP;
793 context->bts_ovfl_signal = cfg.signal;
692 } 794 }
693 795
694 if ((cfg.flags & PTRACE_BTS_O_ALLOC) && 796 ds_release_bts(context->tracer);
695 (cfg.size != child->bts_size)) { 797 context->tracer = NULL;
696 int error;
697 798
698 ptrace_bts_free_buffer(child); 799 if ((cfg.flags & PTRACE_BTS_O_ALLOC) && (cfg.size != context->size)) {
800 int err;
699 801
700 error = ptrace_bts_allocate_buffer(child, cfg.size); 802 free_bts_buffer(context);
701 if (error < 0) 803 if (!cfg.size)
702 return error; 804 return 0;
805
806 err = alloc_bts_buffer(context, cfg.size);
807 if (err < 0)
808 return err;
703 } 809 }
704 810
705 if (cfg.flags & PTRACE_BTS_O_TRACE) 811 if (cfg.flags & PTRACE_BTS_O_TRACE)
@@ -708,15 +814,14 @@ static int ptrace_bts_config(struct task_struct *child,
708 if (cfg.flags & PTRACE_BTS_O_SCHED) 814 if (cfg.flags & PTRACE_BTS_O_SCHED)
709 flags |= BTS_TIMESTAMPS; 815 flags |= BTS_TIMESTAMPS;
710 816
711 child->bts = ds_request_bts(child, child->bts_buffer, child->bts_size, 817 context->tracer =
712 /* ovfl = */ NULL, /* th = */ (size_t)-1, 818 ds_request_bts_task(child, context->buffer, context->size,
713 flags); 819 NULL, (size_t)-1, flags);
714 if (IS_ERR(child->bts)) { 820 if (unlikely(IS_ERR(context->tracer))) {
715 int error = PTR_ERR(child->bts); 821 int error = PTR_ERR(context->tracer);
716
717 ptrace_bts_free_buffer(child);
718 child->bts = NULL;
719 822
823 free_bts_buffer(context);
824 context->tracer = NULL;
720 return error; 825 return error;
721 } 826 }
722 827
@@ -727,20 +832,25 @@ static int ptrace_bts_status(struct task_struct *child,
727 long cfg_size, 832 long cfg_size,
728 struct ptrace_bts_config __user *ucfg) 833 struct ptrace_bts_config __user *ucfg)
729{ 834{
835 struct bts_context *context;
730 const struct bts_trace *trace; 836 const struct bts_trace *trace;
731 struct ptrace_bts_config cfg; 837 struct ptrace_bts_config cfg;
732 838
839 context = child->bts;
840 if (!context)
841 return -ESRCH;
842
733 if (cfg_size < sizeof(cfg)) 843 if (cfg_size < sizeof(cfg))
734 return -EIO; 844 return -EIO;
735 845
736 trace = ds_read_bts(child->bts); 846 trace = ds_read_bts(context->tracer);
737 if (!trace) 847 if (!trace)
738 return -EPERM; 848 return -ESRCH;
739 849
740 memset(&cfg, 0, sizeof(cfg)); 850 memset(&cfg, 0, sizeof(cfg));
741 cfg.size = trace->ds.end - trace->ds.begin; 851 cfg.size = trace->ds.end - trace->ds.begin;
742 cfg.signal = child->thread.bts_ovfl_signal; 852 cfg.signal = context->bts_ovfl_signal;
743 cfg.bts_size = sizeof(struct bts_struct); 853 cfg.bts_size = sizeof(struct bts_struct);
744 854
745 if (cfg.signal) 855 if (cfg.signal)
746 cfg.flags |= PTRACE_BTS_O_SIGNAL; 856 cfg.flags |= PTRACE_BTS_O_SIGNAL;
@@ -759,80 +869,51 @@ static int ptrace_bts_status(struct task_struct *child,
759 869
760static int ptrace_bts_clear(struct task_struct *child) 870static int ptrace_bts_clear(struct task_struct *child)
761{ 871{
872 struct bts_context *context;
762 const struct bts_trace *trace; 873 const struct bts_trace *trace;
763 874
764 trace = ds_read_bts(child->bts); 875 context = child->bts;
876 if (!context)
877 return -ESRCH;
878
879 trace = ds_read_bts(context->tracer);
765 if (!trace) 880 if (!trace)
766 return -EPERM; 881 return -ESRCH;
767 882
768 memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size); 883 memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
769 884
770 return ds_reset_bts(child->bts); 885 return ds_reset_bts(context->tracer);
771} 886}
772 887
773static int ptrace_bts_size(struct task_struct *child) 888static int ptrace_bts_size(struct task_struct *child)
774{ 889{
890 struct bts_context *context;
775 const struct bts_trace *trace; 891 const struct bts_trace *trace;
776 892
777 trace = ds_read_bts(child->bts); 893 context = child->bts;
894 if (!context)
895 return -ESRCH;
896
897 trace = ds_read_bts(context->tracer);
778 if (!trace) 898 if (!trace)
779 return -EPERM; 899 return -ESRCH;
780 900
781 return (trace->ds.top - trace->ds.begin) / trace->ds.size; 901 return (trace->ds.top - trace->ds.begin) / trace->ds.size;
782} 902}
783 903
784static void ptrace_bts_fork(struct task_struct *tsk) 904/*
785{ 905 * Called from __ptrace_unlink() after the child has been moved back
786 tsk->bts = NULL; 906 * to its original parent.
787 tsk->bts_buffer = NULL; 907 */
788 tsk->bts_size = 0; 908void ptrace_bts_untrace(struct task_struct *child)
789 tsk->thread.bts_ovfl_signal = 0;
790}
791
792static void ptrace_bts_untrace(struct task_struct *child)
793{ 909{
794 if (unlikely(child->bts)) { 910 if (unlikely(child->bts)) {
795 ds_release_bts(child->bts); 911 free_bts_context(child->bts);
796 child->bts = NULL; 912 child->bts = NULL;
797
798 /* We cannot update total_vm and locked_vm since
799 child's mm is already gone. But we can reclaim the
800 memory. */
801 kfree(child->bts_buffer);
802 child->bts_buffer = NULL;
803 child->bts_size = 0;
804 } 913 }
805} 914}
806
807static void ptrace_bts_detach(struct task_struct *child)
808{
809 /*
810 * Ptrace_detach() races with ptrace_untrace() in case
811 * the child dies and is reaped by another thread.
812 *
813 * We only do the memory accounting at this point and
814 * leave the buffer deallocation and the bts tracer
815 * release to ptrace_bts_untrace() which will be called
816 * later on with tasklist_lock held.
817 */
818 release_locked_buffer(child->bts_buffer, child->bts_size);
819}
820#else
821static inline void ptrace_bts_fork(struct task_struct *tsk) {}
822static inline void ptrace_bts_detach(struct task_struct *child) {}
823static inline void ptrace_bts_untrace(struct task_struct *child) {}
824#endif /* CONFIG_X86_PTRACE_BTS */ 915#endif /* CONFIG_X86_PTRACE_BTS */
825 916
826void x86_ptrace_fork(struct task_struct *child, unsigned long clone_flags)
827{
828 ptrace_bts_fork(child);
829}
830
831void x86_ptrace_untrace(struct task_struct *child)
832{
833 ptrace_bts_untrace(child);
834}
835
836/* 917/*
837 * Called by kernel/ptrace.c when detaching.. 918 * Called by kernel/ptrace.c when detaching..
838 * 919 *
@@ -844,7 +925,6 @@ void ptrace_disable(struct task_struct *child)
844#ifdef TIF_SYSCALL_EMU 925#ifdef TIF_SYSCALL_EMU
845 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU); 926 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
846#endif 927#endif
847 ptrace_bts_detach(child);
848} 928}
849 929
850#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION 930#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
diff --git a/arch/x86/kernel/stacktrace.c b/arch/x86/kernel/stacktrace.c
index f7bddc2e37d1..4aaf7e48394f 100644
--- a/arch/x86/kernel/stacktrace.c
+++ b/arch/x86/kernel/stacktrace.c
@@ -20,7 +20,7 @@ save_stack_warning_symbol(void *data, char *msg, unsigned long symbol)
20 20
21static int save_stack_stack(void *data, char *name) 21static int save_stack_stack(void *data, char *name)
22{ 22{
23 return -1; 23 return 0;
24} 24}
25 25
26static void save_stack_address(void *data, unsigned long addr, int reliable) 26static void save_stack_address(void *data, unsigned long addr, int reliable)
diff --git a/arch/x86/mm/kmmio.c b/arch/x86/mm/kmmio.c
index 50dc802a1c46..16ccbd77917f 100644
--- a/arch/x86/mm/kmmio.c
+++ b/arch/x86/mm/kmmio.c
@@ -32,7 +32,7 @@ struct kmmio_fault_page {
32 struct list_head list; 32 struct list_head list;
33 struct kmmio_fault_page *release_next; 33 struct kmmio_fault_page *release_next;
34 unsigned long page; /* location of the fault page */ 34 unsigned long page; /* location of the fault page */
35 bool old_presence; /* page presence prior to arming */ 35 pteval_t old_presence; /* page presence prior to arming */
36 bool armed; 36 bool armed;
37 37
38 /* 38 /*
@@ -97,60 +97,62 @@ static struct kmmio_probe *get_kmmio_probe(unsigned long addr)
97static struct kmmio_fault_page *get_kmmio_fault_page(unsigned long page) 97static struct kmmio_fault_page *get_kmmio_fault_page(unsigned long page)
98{ 98{
99 struct list_head *head; 99 struct list_head *head;
100 struct kmmio_fault_page *p; 100 struct kmmio_fault_page *f;
101 101
102 page &= PAGE_MASK; 102 page &= PAGE_MASK;
103 head = kmmio_page_list(page); 103 head = kmmio_page_list(page);
104 list_for_each_entry_rcu(p, head, list) { 104 list_for_each_entry_rcu(f, head, list) {
105 if (p->page == page) 105 if (f->page == page)
106 return p; 106 return f;
107 } 107 }
108 return NULL; 108 return NULL;
109} 109}
110 110
111static void set_pmd_presence(pmd_t *pmd, bool present, bool *old) 111static void clear_pmd_presence(pmd_t *pmd, bool clear, pmdval_t *old)
112{ 112{
113 pmdval_t v = pmd_val(*pmd); 113 pmdval_t v = pmd_val(*pmd);
114 *old = !!(v & _PAGE_PRESENT); 114 if (clear) {
115 v &= ~_PAGE_PRESENT; 115 *old = v & _PAGE_PRESENT;
116 if (present) 116 v &= ~_PAGE_PRESENT;
117 v |= _PAGE_PRESENT; 117 } else /* presume this has been called with clear==true previously */
118 v |= *old;
118 set_pmd(pmd, __pmd(v)); 119 set_pmd(pmd, __pmd(v));
119} 120}
120 121
121static void set_pte_presence(pte_t *pte, bool present, bool *old) 122static void clear_pte_presence(pte_t *pte, bool clear, pteval_t *old)
122{ 123{
123 pteval_t v = pte_val(*pte); 124 pteval_t v = pte_val(*pte);
124 *old = !!(v & _PAGE_PRESENT); 125 if (clear) {
125 v &= ~_PAGE_PRESENT; 126 *old = v & _PAGE_PRESENT;
126 if (present) 127 v &= ~_PAGE_PRESENT;
127 v |= _PAGE_PRESENT; 128 } else /* presume this has been called with clear==true previously */
129 v |= *old;
128 set_pte_atomic(pte, __pte(v)); 130 set_pte_atomic(pte, __pte(v));
129} 131}
130 132
131static int set_page_presence(unsigned long addr, bool present, bool *old) 133static int clear_page_presence(struct kmmio_fault_page *f, bool clear)
132{ 134{
133 unsigned int level; 135 unsigned int level;
134 pte_t *pte = lookup_address(addr, &level); 136 pte_t *pte = lookup_address(f->page, &level);
135 137
136 if (!pte) { 138 if (!pte) {
137 pr_err("kmmio: no pte for page 0x%08lx\n", addr); 139 pr_err("kmmio: no pte for page 0x%08lx\n", f->page);
138 return -1; 140 return -1;
139 } 141 }
140 142
141 switch (level) { 143 switch (level) {
142 case PG_LEVEL_2M: 144 case PG_LEVEL_2M:
143 set_pmd_presence((pmd_t *)pte, present, old); 145 clear_pmd_presence((pmd_t *)pte, clear, &f->old_presence);
144 break; 146 break;
145 case PG_LEVEL_4K: 147 case PG_LEVEL_4K:
146 set_pte_presence(pte, present, old); 148 clear_pte_presence(pte, clear, &f->old_presence);
147 break; 149 break;
148 default: 150 default:
149 pr_err("kmmio: unexpected page level 0x%x.\n", level); 151 pr_err("kmmio: unexpected page level 0x%x.\n", level);
150 return -1; 152 return -1;
151 } 153 }
152 154
153 __flush_tlb_one(addr); 155 __flush_tlb_one(f->page);
154 return 0; 156 return 0;
155} 157}
156 158
@@ -171,9 +173,9 @@ static int arm_kmmio_fault_page(struct kmmio_fault_page *f)
171 WARN_ONCE(f->armed, KERN_ERR "kmmio page already armed.\n"); 173 WARN_ONCE(f->armed, KERN_ERR "kmmio page already armed.\n");
172 if (f->armed) { 174 if (f->armed) {
173 pr_warning("kmmio double-arm: page 0x%08lx, ref %d, old %d\n", 175 pr_warning("kmmio double-arm: page 0x%08lx, ref %d, old %d\n",
174 f->page, f->count, f->old_presence); 176 f->page, f->count, !!f->old_presence);
175 } 177 }
176 ret = set_page_presence(f->page, false, &f->old_presence); 178 ret = clear_page_presence(f, true);
177 WARN_ONCE(ret < 0, KERN_ERR "kmmio arming 0x%08lx failed.\n", f->page); 179 WARN_ONCE(ret < 0, KERN_ERR "kmmio arming 0x%08lx failed.\n", f->page);
178 f->armed = true; 180 f->armed = true;
179 return ret; 181 return ret;
@@ -182,8 +184,7 @@ static int arm_kmmio_fault_page(struct kmmio_fault_page *f)
182/** Restore the given page to saved presence state. */ 184/** Restore the given page to saved presence state. */
183static void disarm_kmmio_fault_page(struct kmmio_fault_page *f) 185static void disarm_kmmio_fault_page(struct kmmio_fault_page *f)
184{ 186{
185 bool tmp; 187 int ret = clear_page_presence(f, false);
186 int ret = set_page_presence(f->page, f->old_presence, &tmp);
187 WARN_ONCE(ret < 0, 188 WARN_ONCE(ret < 0,
188 KERN_ERR "kmmio disarming 0x%08lx failed.\n", f->page); 189 KERN_ERR "kmmio disarming 0x%08lx failed.\n", f->page);
189 f->armed = false; 190 f->armed = false;
@@ -310,7 +311,12 @@ static int post_kmmio_handler(unsigned long condition, struct pt_regs *regs)
310 struct kmmio_context *ctx = &get_cpu_var(kmmio_ctx); 311 struct kmmio_context *ctx = &get_cpu_var(kmmio_ctx);
311 312
312 if (!ctx->active) { 313 if (!ctx->active) {
313 pr_debug("kmmio: spurious debug trap on CPU %d.\n", 314 /*
315 * debug traps without an active context are due to either
316 * something external causing them (f.e. using a debugger while
317 * mmio tracing enabled), or erroneous behaviour
318 */
319 pr_warning("kmmio: unexpected debug trap on CPU %d.\n",
314 smp_processor_id()); 320 smp_processor_id());
315 goto out; 321 goto out;
316 } 322 }
@@ -439,12 +445,12 @@ static void rcu_free_kmmio_fault_pages(struct rcu_head *head)
439 head, 445 head,
440 struct kmmio_delayed_release, 446 struct kmmio_delayed_release,
441 rcu); 447 rcu);
442 struct kmmio_fault_page *p = dr->release_list; 448 struct kmmio_fault_page *f = dr->release_list;
443 while (p) { 449 while (f) {
444 struct kmmio_fault_page *next = p->release_next; 450 struct kmmio_fault_page *next = f->release_next;
445 BUG_ON(p->count); 451 BUG_ON(f->count);
446 kfree(p); 452 kfree(f);
447 p = next; 453 f = next;
448 } 454 }
449 kfree(dr); 455 kfree(dr);
450} 456}
@@ -453,19 +459,19 @@ static void remove_kmmio_fault_pages(struct rcu_head *head)
453{ 459{
454 struct kmmio_delayed_release *dr = 460 struct kmmio_delayed_release *dr =
455 container_of(head, struct kmmio_delayed_release, rcu); 461 container_of(head, struct kmmio_delayed_release, rcu);
456 struct kmmio_fault_page *p = dr->release_list; 462 struct kmmio_fault_page *f = dr->release_list;
457 struct kmmio_fault_page **prevp = &dr->release_list; 463 struct kmmio_fault_page **prevp = &dr->release_list;
458 unsigned long flags; 464 unsigned long flags;
459 465
460 spin_lock_irqsave(&kmmio_lock, flags); 466 spin_lock_irqsave(&kmmio_lock, flags);
461 while (p) { 467 while (f) {
462 if (!p->count) { 468 if (!f->count) {
463 list_del_rcu(&p->list); 469 list_del_rcu(&f->list);
464 prevp = &p->release_next; 470 prevp = &f->release_next;
465 } else { 471 } else {
466 *prevp = p->release_next; 472 *prevp = f->release_next;
467 } 473 }
468 p = p->release_next; 474 f = f->release_next;
469 } 475 }
470 spin_unlock_irqrestore(&kmmio_lock, flags); 476 spin_unlock_irqrestore(&kmmio_lock, flags);
471 477
@@ -528,8 +534,8 @@ void unregister_kmmio_probe(struct kmmio_probe *p)
528} 534}
529EXPORT_SYMBOL(unregister_kmmio_probe); 535EXPORT_SYMBOL(unregister_kmmio_probe);
530 536
531static int kmmio_die_notifier(struct notifier_block *nb, unsigned long val, 537static int
532 void *args) 538kmmio_die_notifier(struct notifier_block *nb, unsigned long val, void *args)
533{ 539{
534 struct die_args *arg = args; 540 struct die_args *arg = args;
535 541
@@ -544,11 +550,23 @@ static struct notifier_block nb_die = {
544 .notifier_call = kmmio_die_notifier 550 .notifier_call = kmmio_die_notifier
545}; 551};
546 552
547static int __init init_kmmio(void) 553int kmmio_init(void)
548{ 554{
549 int i; 555 int i;
556
550 for (i = 0; i < KMMIO_PAGE_TABLE_SIZE; i++) 557 for (i = 0; i < KMMIO_PAGE_TABLE_SIZE; i++)
551 INIT_LIST_HEAD(&kmmio_page_table[i]); 558 INIT_LIST_HEAD(&kmmio_page_table[i]);
559
552 return register_die_notifier(&nb_die); 560 return register_die_notifier(&nb_die);
553} 561}
554fs_initcall(init_kmmio); /* should be before device_initcall() */ 562
563void kmmio_cleanup(void)
564{
565 int i;
566
567 unregister_die_notifier(&nb_die);
568 for (i = 0; i < KMMIO_PAGE_TABLE_SIZE; i++) {
569 WARN_ONCE(!list_empty(&kmmio_page_table[i]),
570 KERN_ERR "kmmio_page_table not empty at cleanup, any further tracing will leak memory.\n");
571 }
572}
diff --git a/arch/x86/mm/mmio-mod.c b/arch/x86/mm/mmio-mod.c
index c9342ed8b402..132772a8ec57 100644
--- a/arch/x86/mm/mmio-mod.c
+++ b/arch/x86/mm/mmio-mod.c
@@ -451,6 +451,7 @@ void enable_mmiotrace(void)
451 451
452 if (nommiotrace) 452 if (nommiotrace)
453 pr_info(NAME "MMIO tracing disabled.\n"); 453 pr_info(NAME "MMIO tracing disabled.\n");
454 kmmio_init();
454 enter_uniprocessor(); 455 enter_uniprocessor();
455 spin_lock_irq(&trace_lock); 456 spin_lock_irq(&trace_lock);
456 atomic_inc(&mmiotrace_enabled); 457 atomic_inc(&mmiotrace_enabled);
@@ -473,6 +474,7 @@ void disable_mmiotrace(void)
473 474
474 clear_trace_list(); /* guarantees: no more kmmio callbacks */ 475 clear_trace_list(); /* guarantees: no more kmmio callbacks */
475 leave_uniprocessor(); 476 leave_uniprocessor();
477 kmmio_cleanup();
476 pr_info(NAME "disabled.\n"); 478 pr_info(NAME "disabled.\n");
477out: 479out:
478 mutex_unlock(&mmiotrace_mutex); 480 mutex_unlock(&mmiotrace_mutex);