diff options
Diffstat (limited to 'arch/x86')
-rw-r--r-- | arch/x86/Kconfig | 2 | ||||
-rw-r--r-- | arch/x86/include/asm/cpufeature.h | 2 | ||||
-rw-r--r-- | arch/x86/include/asm/ftrace.h | 1 | ||||
-rw-r--r-- | arch/x86/include/asm/perf_event.h | 13 | ||||
-rw-r--r-- | arch/x86/include/uapi/asm/msr-index.h | 2 | ||||
-rw-r--r-- | arch/x86/kernel/Makefile | 3 | ||||
-rw-r--r-- | arch/x86/kernel/cpu/perf_event.c | 15 | ||||
-rw-r--r-- | arch/x86/kernel/cpu/perf_event.h | 25 | ||||
-rw-r--r-- | arch/x86/kernel/cpu/perf_event_amd.c | 322 | ||||
-rw-r--r-- | arch/x86/kernel/kprobes/Makefile | 7 | ||||
-rw-r--r-- | arch/x86/kernel/kprobes/common.h (renamed from arch/x86/kernel/kprobes-common.h) | 11 | ||||
-rw-r--r-- | arch/x86/kernel/kprobes/core.c (renamed from arch/x86/kernel/kprobes.c) | 76 | ||||
-rw-r--r-- | arch/x86/kernel/kprobes/ftrace.c | 93 | ||||
-rw-r--r-- | arch/x86/kernel/kprobes/opt.c (renamed from arch/x86/kernel/kprobes-opt.c) | 2 | ||||
-rw-r--r-- | arch/x86/kernel/uprobes.c | 4 |
15 files changed, 398 insertions, 180 deletions
diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig index 36b05ed0bb3c..a2570490eab6 100644 --- a/arch/x86/Kconfig +++ b/arch/x86/Kconfig | |||
@@ -39,10 +39,12 @@ config X86 | |||
39 | select HAVE_DMA_CONTIGUOUS if !SWIOTLB | 39 | select HAVE_DMA_CONTIGUOUS if !SWIOTLB |
40 | select HAVE_KRETPROBES | 40 | select HAVE_KRETPROBES |
41 | select HAVE_OPTPROBES | 41 | select HAVE_OPTPROBES |
42 | select HAVE_KPROBES_ON_FTRACE | ||
42 | select HAVE_FTRACE_MCOUNT_RECORD | 43 | select HAVE_FTRACE_MCOUNT_RECORD |
43 | select HAVE_FENTRY if X86_64 | 44 | select HAVE_FENTRY if X86_64 |
44 | select HAVE_C_RECORDMCOUNT | 45 | select HAVE_C_RECORDMCOUNT |
45 | select HAVE_DYNAMIC_FTRACE | 46 | select HAVE_DYNAMIC_FTRACE |
47 | select HAVE_DYNAMIC_FTRACE_WITH_REGS | ||
46 | select HAVE_FUNCTION_TRACER | 48 | select HAVE_FUNCTION_TRACER |
47 | select HAVE_FUNCTION_GRAPH_TRACER | 49 | select HAVE_FUNCTION_GRAPH_TRACER |
48 | select HAVE_FUNCTION_GRAPH_FP_TEST | 50 | select HAVE_FUNCTION_GRAPH_FP_TEST |
diff --git a/arch/x86/include/asm/cpufeature.h b/arch/x86/include/asm/cpufeature.h index 2d9075e863a0..93fe929d1cee 100644 --- a/arch/x86/include/asm/cpufeature.h +++ b/arch/x86/include/asm/cpufeature.h | |||
@@ -167,6 +167,7 @@ | |||
167 | #define X86_FEATURE_TBM (6*32+21) /* trailing bit manipulations */ | 167 | #define X86_FEATURE_TBM (6*32+21) /* trailing bit manipulations */ |
168 | #define X86_FEATURE_TOPOEXT (6*32+22) /* topology extensions CPUID leafs */ | 168 | #define X86_FEATURE_TOPOEXT (6*32+22) /* topology extensions CPUID leafs */ |
169 | #define X86_FEATURE_PERFCTR_CORE (6*32+23) /* core performance counter extensions */ | 169 | #define X86_FEATURE_PERFCTR_CORE (6*32+23) /* core performance counter extensions */ |
170 | #define X86_FEATURE_PERFCTR_NB (6*32+24) /* NB performance counter extensions */ | ||
170 | 171 | ||
171 | /* | 172 | /* |
172 | * Auxiliary flags: Linux defined - For features scattered in various | 173 | * Auxiliary flags: Linux defined - For features scattered in various |
@@ -309,6 +310,7 @@ extern const char * const x86_power_flags[32]; | |||
309 | #define cpu_has_hypervisor boot_cpu_has(X86_FEATURE_HYPERVISOR) | 310 | #define cpu_has_hypervisor boot_cpu_has(X86_FEATURE_HYPERVISOR) |
310 | #define cpu_has_pclmulqdq boot_cpu_has(X86_FEATURE_PCLMULQDQ) | 311 | #define cpu_has_pclmulqdq boot_cpu_has(X86_FEATURE_PCLMULQDQ) |
311 | #define cpu_has_perfctr_core boot_cpu_has(X86_FEATURE_PERFCTR_CORE) | 312 | #define cpu_has_perfctr_core boot_cpu_has(X86_FEATURE_PERFCTR_CORE) |
313 | #define cpu_has_perfctr_nb boot_cpu_has(X86_FEATURE_PERFCTR_NB) | ||
312 | #define cpu_has_cx8 boot_cpu_has(X86_FEATURE_CX8) | 314 | #define cpu_has_cx8 boot_cpu_has(X86_FEATURE_CX8) |
313 | #define cpu_has_cx16 boot_cpu_has(X86_FEATURE_CX16) | 315 | #define cpu_has_cx16 boot_cpu_has(X86_FEATURE_CX16) |
314 | #define cpu_has_eager_fpu boot_cpu_has(X86_FEATURE_EAGER_FPU) | 316 | #define cpu_has_eager_fpu boot_cpu_has(X86_FEATURE_EAGER_FPU) |
diff --git a/arch/x86/include/asm/ftrace.h b/arch/x86/include/asm/ftrace.h index 9a25b522d377..86cb51e1ca96 100644 --- a/arch/x86/include/asm/ftrace.h +++ b/arch/x86/include/asm/ftrace.h | |||
@@ -44,7 +44,6 @@ | |||
44 | 44 | ||
45 | #ifdef CONFIG_DYNAMIC_FTRACE | 45 | #ifdef CONFIG_DYNAMIC_FTRACE |
46 | #define ARCH_SUPPORTS_FTRACE_OPS 1 | 46 | #define ARCH_SUPPORTS_FTRACE_OPS 1 |
47 | #define ARCH_SUPPORTS_FTRACE_SAVE_REGS | ||
48 | #endif | 47 | #endif |
49 | 48 | ||
50 | #ifndef __ASSEMBLY__ | 49 | #ifndef __ASSEMBLY__ |
diff --git a/arch/x86/include/asm/perf_event.h b/arch/x86/include/asm/perf_event.h index 4fabcdf1cfa7..57cb63402213 100644 --- a/arch/x86/include/asm/perf_event.h +++ b/arch/x86/include/asm/perf_event.h | |||
@@ -29,8 +29,13 @@ | |||
29 | #define ARCH_PERFMON_EVENTSEL_INV (1ULL << 23) | 29 | #define ARCH_PERFMON_EVENTSEL_INV (1ULL << 23) |
30 | #define ARCH_PERFMON_EVENTSEL_CMASK 0xFF000000ULL | 30 | #define ARCH_PERFMON_EVENTSEL_CMASK 0xFF000000ULL |
31 | 31 | ||
32 | #define AMD_PERFMON_EVENTSEL_GUESTONLY (1ULL << 40) | 32 | #define AMD64_EVENTSEL_INT_CORE_ENABLE (1ULL << 36) |
33 | #define AMD_PERFMON_EVENTSEL_HOSTONLY (1ULL << 41) | 33 | #define AMD64_EVENTSEL_GUESTONLY (1ULL << 40) |
34 | #define AMD64_EVENTSEL_HOSTONLY (1ULL << 41) | ||
35 | |||
36 | #define AMD64_EVENTSEL_INT_CORE_SEL_SHIFT 37 | ||
37 | #define AMD64_EVENTSEL_INT_CORE_SEL_MASK \ | ||
38 | (0xFULL << AMD64_EVENTSEL_INT_CORE_SEL_SHIFT) | ||
34 | 39 | ||
35 | #define AMD64_EVENTSEL_EVENT \ | 40 | #define AMD64_EVENTSEL_EVENT \ |
36 | (ARCH_PERFMON_EVENTSEL_EVENT | (0x0FULL << 32)) | 41 | (ARCH_PERFMON_EVENTSEL_EVENT | (0x0FULL << 32)) |
@@ -46,8 +51,12 @@ | |||
46 | #define AMD64_RAW_EVENT_MASK \ | 51 | #define AMD64_RAW_EVENT_MASK \ |
47 | (X86_RAW_EVENT_MASK | \ | 52 | (X86_RAW_EVENT_MASK | \ |
48 | AMD64_EVENTSEL_EVENT) | 53 | AMD64_EVENTSEL_EVENT) |
54 | #define AMD64_RAW_EVENT_MASK_NB \ | ||
55 | (AMD64_EVENTSEL_EVENT | \ | ||
56 | ARCH_PERFMON_EVENTSEL_UMASK) | ||
49 | #define AMD64_NUM_COUNTERS 4 | 57 | #define AMD64_NUM_COUNTERS 4 |
50 | #define AMD64_NUM_COUNTERS_CORE 6 | 58 | #define AMD64_NUM_COUNTERS_CORE 6 |
59 | #define AMD64_NUM_COUNTERS_NB 4 | ||
51 | 60 | ||
52 | #define ARCH_PERFMON_UNHALTED_CORE_CYCLES_SEL 0x3c | 61 | #define ARCH_PERFMON_UNHALTED_CORE_CYCLES_SEL 0x3c |
53 | #define ARCH_PERFMON_UNHALTED_CORE_CYCLES_UMASK (0x00 << 8) | 62 | #define ARCH_PERFMON_UNHALTED_CORE_CYCLES_UMASK (0x00 << 8) |
diff --git a/arch/x86/include/uapi/asm/msr-index.h b/arch/x86/include/uapi/asm/msr-index.h index 433a59fb1a74..075a40255591 100644 --- a/arch/x86/include/uapi/asm/msr-index.h +++ b/arch/x86/include/uapi/asm/msr-index.h | |||
@@ -194,6 +194,8 @@ | |||
194 | /* Fam 15h MSRs */ | 194 | /* Fam 15h MSRs */ |
195 | #define MSR_F15H_PERF_CTL 0xc0010200 | 195 | #define MSR_F15H_PERF_CTL 0xc0010200 |
196 | #define MSR_F15H_PERF_CTR 0xc0010201 | 196 | #define MSR_F15H_PERF_CTR 0xc0010201 |
197 | #define MSR_F15H_NB_PERF_CTL 0xc0010240 | ||
198 | #define MSR_F15H_NB_PERF_CTR 0xc0010241 | ||
197 | 199 | ||
198 | /* Fam 10h MSRs */ | 200 | /* Fam 10h MSRs */ |
199 | #define MSR_FAM10H_MMIO_CONF_BASE 0xc0010058 | 201 | #define MSR_FAM10H_MMIO_CONF_BASE 0xc0010058 |
diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile index 34e923a53762..ac3b3d002833 100644 --- a/arch/x86/kernel/Makefile +++ b/arch/x86/kernel/Makefile | |||
@@ -65,8 +65,7 @@ obj-$(CONFIG_X86_TSC) += trace_clock.o | |||
65 | obj-$(CONFIG_KEXEC) += machine_kexec_$(BITS).o | 65 | obj-$(CONFIG_KEXEC) += machine_kexec_$(BITS).o |
66 | obj-$(CONFIG_KEXEC) += relocate_kernel_$(BITS).o crash.o | 66 | obj-$(CONFIG_KEXEC) += relocate_kernel_$(BITS).o crash.o |
67 | obj-$(CONFIG_CRASH_DUMP) += crash_dump_$(BITS).o | 67 | obj-$(CONFIG_CRASH_DUMP) += crash_dump_$(BITS).o |
68 | obj-$(CONFIG_KPROBES) += kprobes.o | 68 | obj-y += kprobes/ |
69 | obj-$(CONFIG_OPTPROBES) += kprobes-opt.o | ||
70 | obj-$(CONFIG_MODULES) += module.o | 69 | obj-$(CONFIG_MODULES) += module.o |
71 | obj-$(CONFIG_DOUBLEFAULT) += doublefault_32.o | 70 | obj-$(CONFIG_DOUBLEFAULT) += doublefault_32.o |
72 | obj-$(CONFIG_KGDB) += kgdb.o | 71 | obj-$(CONFIG_KGDB) += kgdb.o |
diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c index 6774c17a5576..bf0f01aea994 100644 --- a/arch/x86/kernel/cpu/perf_event.c +++ b/arch/x86/kernel/cpu/perf_event.c | |||
@@ -829,7 +829,7 @@ static inline void x86_assign_hw_event(struct perf_event *event, | |||
829 | } else { | 829 | } else { |
830 | hwc->config_base = x86_pmu_config_addr(hwc->idx); | 830 | hwc->config_base = x86_pmu_config_addr(hwc->idx); |
831 | hwc->event_base = x86_pmu_event_addr(hwc->idx); | 831 | hwc->event_base = x86_pmu_event_addr(hwc->idx); |
832 | hwc->event_base_rdpmc = hwc->idx; | 832 | hwc->event_base_rdpmc = x86_pmu_rdpmc_index(hwc->idx); |
833 | } | 833 | } |
834 | } | 834 | } |
835 | 835 | ||
@@ -1310,11 +1310,6 @@ static struct attribute_group x86_pmu_format_group = { | |||
1310 | .attrs = NULL, | 1310 | .attrs = NULL, |
1311 | }; | 1311 | }; |
1312 | 1312 | ||
1313 | struct perf_pmu_events_attr { | ||
1314 | struct device_attribute attr; | ||
1315 | u64 id; | ||
1316 | }; | ||
1317 | |||
1318 | /* | 1313 | /* |
1319 | * Remove all undefined events (x86_pmu.event_map(id) == 0) | 1314 | * Remove all undefined events (x86_pmu.event_map(id) == 0) |
1320 | * out of events_attr attributes. | 1315 | * out of events_attr attributes. |
@@ -1348,11 +1343,9 @@ static ssize_t events_sysfs_show(struct device *dev, struct device_attribute *at | |||
1348 | #define EVENT_VAR(_id) event_attr_##_id | 1343 | #define EVENT_VAR(_id) event_attr_##_id |
1349 | #define EVENT_PTR(_id) &event_attr_##_id.attr.attr | 1344 | #define EVENT_PTR(_id) &event_attr_##_id.attr.attr |
1350 | 1345 | ||
1351 | #define EVENT_ATTR(_name, _id) \ | 1346 | #define EVENT_ATTR(_name, _id) \ |
1352 | static struct perf_pmu_events_attr EVENT_VAR(_id) = { \ | 1347 | PMU_EVENT_ATTR(_name, EVENT_VAR(_id), PERF_COUNT_HW_##_id, \ |
1353 | .attr = __ATTR(_name, 0444, events_sysfs_show, NULL), \ | 1348 | events_sysfs_show) |
1354 | .id = PERF_COUNT_HW_##_id, \ | ||
1355 | }; | ||
1356 | 1349 | ||
1357 | EVENT_ATTR(cpu-cycles, CPU_CYCLES ); | 1350 | EVENT_ATTR(cpu-cycles, CPU_CYCLES ); |
1358 | EVENT_ATTR(instructions, INSTRUCTIONS ); | 1351 | EVENT_ATTR(instructions, INSTRUCTIONS ); |
diff --git a/arch/x86/kernel/cpu/perf_event.h b/arch/x86/kernel/cpu/perf_event.h index 115c1ea97746..7f5c75c2afdd 100644 --- a/arch/x86/kernel/cpu/perf_event.h +++ b/arch/x86/kernel/cpu/perf_event.h | |||
@@ -325,6 +325,8 @@ struct x86_pmu { | |||
325 | int (*schedule_events)(struct cpu_hw_events *cpuc, int n, int *assign); | 325 | int (*schedule_events)(struct cpu_hw_events *cpuc, int n, int *assign); |
326 | unsigned eventsel; | 326 | unsigned eventsel; |
327 | unsigned perfctr; | 327 | unsigned perfctr; |
328 | int (*addr_offset)(int index, bool eventsel); | ||
329 | int (*rdpmc_index)(int index); | ||
328 | u64 (*event_map)(int); | 330 | u64 (*event_map)(int); |
329 | int max_events; | 331 | int max_events; |
330 | int num_counters; | 332 | int num_counters; |
@@ -446,28 +448,21 @@ extern u64 __read_mostly hw_cache_extra_regs | |||
446 | 448 | ||
447 | u64 x86_perf_event_update(struct perf_event *event); | 449 | u64 x86_perf_event_update(struct perf_event *event); |
448 | 450 | ||
449 | static inline int x86_pmu_addr_offset(int index) | 451 | static inline unsigned int x86_pmu_config_addr(int index) |
450 | { | 452 | { |
451 | int offset; | 453 | return x86_pmu.eventsel + (x86_pmu.addr_offset ? |
452 | 454 | x86_pmu.addr_offset(index, true) : index); | |
453 | /* offset = X86_FEATURE_PERFCTR_CORE ? index << 1 : index */ | ||
454 | alternative_io(ASM_NOP2, | ||
455 | "shll $1, %%eax", | ||
456 | X86_FEATURE_PERFCTR_CORE, | ||
457 | "=a" (offset), | ||
458 | "a" (index)); | ||
459 | |||
460 | return offset; | ||
461 | } | 455 | } |
462 | 456 | ||
463 | static inline unsigned int x86_pmu_config_addr(int index) | 457 | static inline unsigned int x86_pmu_event_addr(int index) |
464 | { | 458 | { |
465 | return x86_pmu.eventsel + x86_pmu_addr_offset(index); | 459 | return x86_pmu.perfctr + (x86_pmu.addr_offset ? |
460 | x86_pmu.addr_offset(index, false) : index); | ||
466 | } | 461 | } |
467 | 462 | ||
468 | static inline unsigned int x86_pmu_event_addr(int index) | 463 | static inline int x86_pmu_rdpmc_index(int index) |
469 | { | 464 | { |
470 | return x86_pmu.perfctr + x86_pmu_addr_offset(index); | 465 | return x86_pmu.rdpmc_index ? x86_pmu.rdpmc_index(index) : index; |
471 | } | 466 | } |
472 | 467 | ||
473 | int x86_setup_perfctr(struct perf_event *event); | 468 | int x86_setup_perfctr(struct perf_event *event); |
diff --git a/arch/x86/kernel/cpu/perf_event_amd.c b/arch/x86/kernel/cpu/perf_event_amd.c index c93bc4e813a0..dfdab42aed27 100644 --- a/arch/x86/kernel/cpu/perf_event_amd.c +++ b/arch/x86/kernel/cpu/perf_event_amd.c | |||
@@ -132,21 +132,102 @@ static u64 amd_pmu_event_map(int hw_event) | |||
132 | return amd_perfmon_event_map[hw_event]; | 132 | return amd_perfmon_event_map[hw_event]; |
133 | } | 133 | } |
134 | 134 | ||
135 | static int amd_pmu_hw_config(struct perf_event *event) | 135 | static struct event_constraint *amd_nb_event_constraint; |
136 | |||
137 | /* | ||
138 | * Previously calculated offsets | ||
139 | */ | ||
140 | static unsigned int event_offsets[X86_PMC_IDX_MAX] __read_mostly; | ||
141 | static unsigned int count_offsets[X86_PMC_IDX_MAX] __read_mostly; | ||
142 | static unsigned int rdpmc_indexes[X86_PMC_IDX_MAX] __read_mostly; | ||
143 | |||
144 | /* | ||
145 | * Legacy CPUs: | ||
146 | * 4 counters starting at 0xc0010000 each offset by 1 | ||
147 | * | ||
148 | * CPUs with core performance counter extensions: | ||
149 | * 6 counters starting at 0xc0010200 each offset by 2 | ||
150 | * | ||
151 | * CPUs with north bridge performance counter extensions: | ||
152 | * 4 additional counters starting at 0xc0010240 each offset by 2 | ||
153 | * (indexed right above either one of the above core counters) | ||
154 | */ | ||
155 | static inline int amd_pmu_addr_offset(int index, bool eventsel) | ||
136 | { | 156 | { |
137 | int ret; | 157 | int offset, first, base; |
138 | 158 | ||
139 | /* pass precise event sampling to ibs: */ | 159 | if (!index) |
140 | if (event->attr.precise_ip && get_ibs_caps()) | 160 | return index; |
141 | return -ENOENT; | 161 | |
162 | if (eventsel) | ||
163 | offset = event_offsets[index]; | ||
164 | else | ||
165 | offset = count_offsets[index]; | ||
166 | |||
167 | if (offset) | ||
168 | return offset; | ||
169 | |||
170 | if (amd_nb_event_constraint && | ||
171 | test_bit(index, amd_nb_event_constraint->idxmsk)) { | ||
172 | /* | ||
173 | * calculate the offset of NB counters with respect to | ||
174 | * base eventsel or perfctr | ||
175 | */ | ||
176 | |||
177 | first = find_first_bit(amd_nb_event_constraint->idxmsk, | ||
178 | X86_PMC_IDX_MAX); | ||
179 | |||
180 | if (eventsel) | ||
181 | base = MSR_F15H_NB_PERF_CTL - x86_pmu.eventsel; | ||
182 | else | ||
183 | base = MSR_F15H_NB_PERF_CTR - x86_pmu.perfctr; | ||
184 | |||
185 | offset = base + ((index - first) << 1); | ||
186 | } else if (!cpu_has_perfctr_core) | ||
187 | offset = index; | ||
188 | else | ||
189 | offset = index << 1; | ||
190 | |||
191 | if (eventsel) | ||
192 | event_offsets[index] = offset; | ||
193 | else | ||
194 | count_offsets[index] = offset; | ||
195 | |||
196 | return offset; | ||
197 | } | ||
198 | |||
199 | static inline int amd_pmu_rdpmc_index(int index) | ||
200 | { | ||
201 | int ret, first; | ||
202 | |||
203 | if (!index) | ||
204 | return index; | ||
205 | |||
206 | ret = rdpmc_indexes[index]; | ||
142 | 207 | ||
143 | ret = x86_pmu_hw_config(event); | ||
144 | if (ret) | 208 | if (ret) |
145 | return ret; | 209 | return ret; |
146 | 210 | ||
147 | if (has_branch_stack(event)) | 211 | if (amd_nb_event_constraint && |
148 | return -EOPNOTSUPP; | 212 | test_bit(index, amd_nb_event_constraint->idxmsk)) { |
213 | /* | ||
214 | * according to the mnual, ECX value of the NB counters is | ||
215 | * the index of the NB counter (0, 1, 2 or 3) plus 6 | ||
216 | */ | ||
217 | |||
218 | first = find_first_bit(amd_nb_event_constraint->idxmsk, | ||
219 | X86_PMC_IDX_MAX); | ||
220 | ret = index - first + 6; | ||
221 | } else | ||
222 | ret = index; | ||
223 | |||
224 | rdpmc_indexes[index] = ret; | ||
225 | |||
226 | return ret; | ||
227 | } | ||
149 | 228 | ||
229 | static int amd_core_hw_config(struct perf_event *event) | ||
230 | { | ||
150 | if (event->attr.exclude_host && event->attr.exclude_guest) | 231 | if (event->attr.exclude_host && event->attr.exclude_guest) |
151 | /* | 232 | /* |
152 | * When HO == GO == 1 the hardware treats that as GO == HO == 0 | 233 | * When HO == GO == 1 the hardware treats that as GO == HO == 0 |
@@ -156,14 +237,37 @@ static int amd_pmu_hw_config(struct perf_event *event) | |||
156 | event->hw.config &= ~(ARCH_PERFMON_EVENTSEL_USR | | 237 | event->hw.config &= ~(ARCH_PERFMON_EVENTSEL_USR | |
157 | ARCH_PERFMON_EVENTSEL_OS); | 238 | ARCH_PERFMON_EVENTSEL_OS); |
158 | else if (event->attr.exclude_host) | 239 | else if (event->attr.exclude_host) |
159 | event->hw.config |= AMD_PERFMON_EVENTSEL_GUESTONLY; | 240 | event->hw.config |= AMD64_EVENTSEL_GUESTONLY; |
160 | else if (event->attr.exclude_guest) | 241 | else if (event->attr.exclude_guest) |
161 | event->hw.config |= AMD_PERFMON_EVENTSEL_HOSTONLY; | 242 | event->hw.config |= AMD64_EVENTSEL_HOSTONLY; |
243 | |||
244 | return 0; | ||
245 | } | ||
246 | |||
247 | /* | ||
248 | * NB counters do not support the following event select bits: | ||
249 | * Host/Guest only | ||
250 | * Counter mask | ||
251 | * Invert counter mask | ||
252 | * Edge detect | ||
253 | * OS/User mode | ||
254 | */ | ||
255 | static int amd_nb_hw_config(struct perf_event *event) | ||
256 | { | ||
257 | /* for NB, we only allow system wide counting mode */ | ||
258 | if (is_sampling_event(event) || event->attach_state & PERF_ATTACH_TASK) | ||
259 | return -EINVAL; | ||
260 | |||
261 | if (event->attr.exclude_user || event->attr.exclude_kernel || | ||
262 | event->attr.exclude_host || event->attr.exclude_guest) | ||
263 | return -EINVAL; | ||
162 | 264 | ||
163 | if (event->attr.type != PERF_TYPE_RAW) | 265 | event->hw.config &= ~(ARCH_PERFMON_EVENTSEL_USR | |
164 | return 0; | 266 | ARCH_PERFMON_EVENTSEL_OS); |
165 | 267 | ||
166 | event->hw.config |= event->attr.config & AMD64_RAW_EVENT_MASK; | 268 | if (event->hw.config & ~(AMD64_RAW_EVENT_MASK_NB | |
269 | ARCH_PERFMON_EVENTSEL_INT)) | ||
270 | return -EINVAL; | ||
167 | 271 | ||
168 | return 0; | 272 | return 0; |
169 | } | 273 | } |
@@ -181,6 +285,11 @@ static inline int amd_is_nb_event(struct hw_perf_event *hwc) | |||
181 | return (hwc->config & 0xe0) == 0xe0; | 285 | return (hwc->config & 0xe0) == 0xe0; |
182 | } | 286 | } |
183 | 287 | ||
288 | static inline int amd_is_perfctr_nb_event(struct hw_perf_event *hwc) | ||
289 | { | ||
290 | return amd_nb_event_constraint && amd_is_nb_event(hwc); | ||
291 | } | ||
292 | |||
184 | static inline int amd_has_nb(struct cpu_hw_events *cpuc) | 293 | static inline int amd_has_nb(struct cpu_hw_events *cpuc) |
185 | { | 294 | { |
186 | struct amd_nb *nb = cpuc->amd_nb; | 295 | struct amd_nb *nb = cpuc->amd_nb; |
@@ -188,20 +297,37 @@ static inline int amd_has_nb(struct cpu_hw_events *cpuc) | |||
188 | return nb && nb->nb_id != -1; | 297 | return nb && nb->nb_id != -1; |
189 | } | 298 | } |
190 | 299 | ||
191 | static void amd_put_event_constraints(struct cpu_hw_events *cpuc, | 300 | static int amd_pmu_hw_config(struct perf_event *event) |
192 | struct perf_event *event) | 301 | { |
302 | int ret; | ||
303 | |||
304 | /* pass precise event sampling to ibs: */ | ||
305 | if (event->attr.precise_ip && get_ibs_caps()) | ||
306 | return -ENOENT; | ||
307 | |||
308 | if (has_branch_stack(event)) | ||
309 | return -EOPNOTSUPP; | ||
310 | |||
311 | ret = x86_pmu_hw_config(event); | ||
312 | if (ret) | ||
313 | return ret; | ||
314 | |||
315 | if (event->attr.type == PERF_TYPE_RAW) | ||
316 | event->hw.config |= event->attr.config & AMD64_RAW_EVENT_MASK; | ||
317 | |||
318 | if (amd_is_perfctr_nb_event(&event->hw)) | ||
319 | return amd_nb_hw_config(event); | ||
320 | |||
321 | return amd_core_hw_config(event); | ||
322 | } | ||
323 | |||
324 | static void __amd_put_nb_event_constraints(struct cpu_hw_events *cpuc, | ||
325 | struct perf_event *event) | ||
193 | { | 326 | { |
194 | struct hw_perf_event *hwc = &event->hw; | ||
195 | struct amd_nb *nb = cpuc->amd_nb; | 327 | struct amd_nb *nb = cpuc->amd_nb; |
196 | int i; | 328 | int i; |
197 | 329 | ||
198 | /* | 330 | /* |
199 | * only care about NB events | ||
200 | */ | ||
201 | if (!(amd_has_nb(cpuc) && amd_is_nb_event(hwc))) | ||
202 | return; | ||
203 | |||
204 | /* | ||
205 | * need to scan whole list because event may not have | 331 | * need to scan whole list because event may not have |
206 | * been assigned during scheduling | 332 | * been assigned during scheduling |
207 | * | 333 | * |
@@ -215,6 +341,19 @@ static void amd_put_event_constraints(struct cpu_hw_events *cpuc, | |||
215 | } | 341 | } |
216 | } | 342 | } |
217 | 343 | ||
344 | static void amd_nb_interrupt_hw_config(struct hw_perf_event *hwc) | ||
345 | { | ||
346 | int core_id = cpu_data(smp_processor_id()).cpu_core_id; | ||
347 | |||
348 | /* deliver interrupts only to this core */ | ||
349 | if (hwc->config & ARCH_PERFMON_EVENTSEL_INT) { | ||
350 | hwc->config |= AMD64_EVENTSEL_INT_CORE_ENABLE; | ||
351 | hwc->config &= ~AMD64_EVENTSEL_INT_CORE_SEL_MASK; | ||
352 | hwc->config |= (u64)(core_id) << | ||
353 | AMD64_EVENTSEL_INT_CORE_SEL_SHIFT; | ||
354 | } | ||
355 | } | ||
356 | |||
218 | /* | 357 | /* |
219 | * AMD64 NorthBridge events need special treatment because | 358 | * AMD64 NorthBridge events need special treatment because |
220 | * counter access needs to be synchronized across all cores | 359 | * counter access needs to be synchronized across all cores |
@@ -247,24 +386,24 @@ static void amd_put_event_constraints(struct cpu_hw_events *cpuc, | |||
247 | * | 386 | * |
248 | * Given that resources are allocated (cmpxchg), they must be | 387 | * Given that resources are allocated (cmpxchg), they must be |
249 | * eventually freed for others to use. This is accomplished by | 388 | * eventually freed for others to use. This is accomplished by |
250 | * calling amd_put_event_constraints(). | 389 | * calling __amd_put_nb_event_constraints() |
251 | * | 390 | * |
252 | * Non NB events are not impacted by this restriction. | 391 | * Non NB events are not impacted by this restriction. |
253 | */ | 392 | */ |
254 | static struct event_constraint * | 393 | static struct event_constraint * |
255 | amd_get_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event) | 394 | __amd_get_nb_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event, |
395 | struct event_constraint *c) | ||
256 | { | 396 | { |
257 | struct hw_perf_event *hwc = &event->hw; | 397 | struct hw_perf_event *hwc = &event->hw; |
258 | struct amd_nb *nb = cpuc->amd_nb; | 398 | struct amd_nb *nb = cpuc->amd_nb; |
259 | struct perf_event *old = NULL; | 399 | struct perf_event *old; |
260 | int max = x86_pmu.num_counters; | 400 | int idx, new = -1; |
261 | int i, j, k = -1; | ||
262 | 401 | ||
263 | /* | 402 | if (!c) |
264 | * if not NB event or no NB, then no constraints | 403 | c = &unconstrained; |
265 | */ | 404 | |
266 | if (!(amd_has_nb(cpuc) && amd_is_nb_event(hwc))) | 405 | if (cpuc->is_fake) |
267 | return &unconstrained; | 406 | return c; |
268 | 407 | ||
269 | /* | 408 | /* |
270 | * detect if already present, if so reuse | 409 | * detect if already present, if so reuse |
@@ -276,48 +415,36 @@ amd_get_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event) | |||
276 | * because of successive calls to x86_schedule_events() from | 415 | * because of successive calls to x86_schedule_events() from |
277 | * hw_perf_group_sched_in() without hw_perf_enable() | 416 | * hw_perf_group_sched_in() without hw_perf_enable() |
278 | */ | 417 | */ |
279 | for (i = 0; i < max; i++) { | 418 | for_each_set_bit(idx, c->idxmsk, x86_pmu.num_counters) { |
280 | /* | 419 | if (new == -1 || hwc->idx == idx) |
281 | * keep track of first free slot | 420 | /* assign free slot, prefer hwc->idx */ |
282 | */ | 421 | old = cmpxchg(nb->owners + idx, NULL, event); |
283 | if (k == -1 && !nb->owners[i]) | 422 | else if (nb->owners[idx] == event) |
284 | k = i; | 423 | /* event already present */ |
424 | old = event; | ||
425 | else | ||
426 | continue; | ||
427 | |||
428 | if (old && old != event) | ||
429 | continue; | ||
430 | |||
431 | /* reassign to this slot */ | ||
432 | if (new != -1) | ||
433 | cmpxchg(nb->owners + new, event, NULL); | ||
434 | new = idx; | ||
285 | 435 | ||
286 | /* already present, reuse */ | 436 | /* already present, reuse */ |
287 | if (nb->owners[i] == event) | 437 | if (old == event) |
288 | goto done; | ||
289 | } | ||
290 | /* | ||
291 | * not present, so grab a new slot | ||
292 | * starting either at: | ||
293 | */ | ||
294 | if (hwc->idx != -1) { | ||
295 | /* previous assignment */ | ||
296 | i = hwc->idx; | ||
297 | } else if (k != -1) { | ||
298 | /* start from free slot found */ | ||
299 | i = k; | ||
300 | } else { | ||
301 | /* | ||
302 | * event not found, no slot found in | ||
303 | * first pass, try again from the | ||
304 | * beginning | ||
305 | */ | ||
306 | i = 0; | ||
307 | } | ||
308 | j = i; | ||
309 | do { | ||
310 | old = cmpxchg(nb->owners+i, NULL, event); | ||
311 | if (!old) | ||
312 | break; | 438 | break; |
313 | if (++i == max) | 439 | } |
314 | i = 0; | 440 | |
315 | } while (i != j); | 441 | if (new == -1) |
316 | done: | 442 | return &emptyconstraint; |
317 | if (!old) | 443 | |
318 | return &nb->event_constraints[i]; | 444 | if (amd_is_perfctr_nb_event(hwc)) |
319 | 445 | amd_nb_interrupt_hw_config(hwc); | |
320 | return &emptyconstraint; | 446 | |
447 | return &nb->event_constraints[new]; | ||
321 | } | 448 | } |
322 | 449 | ||
323 | static struct amd_nb *amd_alloc_nb(int cpu) | 450 | static struct amd_nb *amd_alloc_nb(int cpu) |
@@ -364,7 +491,7 @@ static void amd_pmu_cpu_starting(int cpu) | |||
364 | struct amd_nb *nb; | 491 | struct amd_nb *nb; |
365 | int i, nb_id; | 492 | int i, nb_id; |
366 | 493 | ||
367 | cpuc->perf_ctr_virt_mask = AMD_PERFMON_EVENTSEL_HOSTONLY; | 494 | cpuc->perf_ctr_virt_mask = AMD64_EVENTSEL_HOSTONLY; |
368 | 495 | ||
369 | if (boot_cpu_data.x86_max_cores < 2) | 496 | if (boot_cpu_data.x86_max_cores < 2) |
370 | return; | 497 | return; |
@@ -407,6 +534,26 @@ static void amd_pmu_cpu_dead(int cpu) | |||
407 | } | 534 | } |
408 | } | 535 | } |
409 | 536 | ||
537 | static struct event_constraint * | ||
538 | amd_get_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event) | ||
539 | { | ||
540 | /* | ||
541 | * if not NB event or no NB, then no constraints | ||
542 | */ | ||
543 | if (!(amd_has_nb(cpuc) && amd_is_nb_event(&event->hw))) | ||
544 | return &unconstrained; | ||
545 | |||
546 | return __amd_get_nb_event_constraints(cpuc, event, | ||
547 | amd_nb_event_constraint); | ||
548 | } | ||
549 | |||
550 | static void amd_put_event_constraints(struct cpu_hw_events *cpuc, | ||
551 | struct perf_event *event) | ||
552 | { | ||
553 | if (amd_has_nb(cpuc) && amd_is_nb_event(&event->hw)) | ||
554 | __amd_put_nb_event_constraints(cpuc, event); | ||
555 | } | ||
556 | |||
410 | PMU_FORMAT_ATTR(event, "config:0-7,32-35"); | 557 | PMU_FORMAT_ATTR(event, "config:0-7,32-35"); |
411 | PMU_FORMAT_ATTR(umask, "config:8-15" ); | 558 | PMU_FORMAT_ATTR(umask, "config:8-15" ); |
412 | PMU_FORMAT_ATTR(edge, "config:18" ); | 559 | PMU_FORMAT_ATTR(edge, "config:18" ); |
@@ -496,6 +643,9 @@ static struct event_constraint amd_f15_PMC30 = EVENT_CONSTRAINT_OVERLAP(0, 0x09, | |||
496 | static struct event_constraint amd_f15_PMC50 = EVENT_CONSTRAINT(0, 0x3F, 0); | 643 | static struct event_constraint amd_f15_PMC50 = EVENT_CONSTRAINT(0, 0x3F, 0); |
497 | static struct event_constraint amd_f15_PMC53 = EVENT_CONSTRAINT(0, 0x38, 0); | 644 | static struct event_constraint amd_f15_PMC53 = EVENT_CONSTRAINT(0, 0x38, 0); |
498 | 645 | ||
646 | static struct event_constraint amd_NBPMC96 = EVENT_CONSTRAINT(0, 0x3C0, 0); | ||
647 | static struct event_constraint amd_NBPMC74 = EVENT_CONSTRAINT(0, 0xF0, 0); | ||
648 | |||
499 | static struct event_constraint * | 649 | static struct event_constraint * |
500 | amd_get_event_constraints_f15h(struct cpu_hw_events *cpuc, struct perf_event *event) | 650 | amd_get_event_constraints_f15h(struct cpu_hw_events *cpuc, struct perf_event *event) |
501 | { | 651 | { |
@@ -561,8 +711,8 @@ amd_get_event_constraints_f15h(struct cpu_hw_events *cpuc, struct perf_event *ev | |||
561 | return &amd_f15_PMC20; | 711 | return &amd_f15_PMC20; |
562 | } | 712 | } |
563 | case AMD_EVENT_NB: | 713 | case AMD_EVENT_NB: |
564 | /* not yet implemented */ | 714 | return __amd_get_nb_event_constraints(cpuc, event, |
565 | return &emptyconstraint; | 715 | amd_nb_event_constraint); |
566 | default: | 716 | default: |
567 | return &emptyconstraint; | 717 | return &emptyconstraint; |
568 | } | 718 | } |
@@ -587,6 +737,8 @@ static __initconst const struct x86_pmu amd_pmu = { | |||
587 | .schedule_events = x86_schedule_events, | 737 | .schedule_events = x86_schedule_events, |
588 | .eventsel = MSR_K7_EVNTSEL0, | 738 | .eventsel = MSR_K7_EVNTSEL0, |
589 | .perfctr = MSR_K7_PERFCTR0, | 739 | .perfctr = MSR_K7_PERFCTR0, |
740 | .addr_offset = amd_pmu_addr_offset, | ||
741 | .rdpmc_index = amd_pmu_rdpmc_index, | ||
590 | .event_map = amd_pmu_event_map, | 742 | .event_map = amd_pmu_event_map, |
591 | .max_events = ARRAY_SIZE(amd_perfmon_event_map), | 743 | .max_events = ARRAY_SIZE(amd_perfmon_event_map), |
592 | .num_counters = AMD64_NUM_COUNTERS, | 744 | .num_counters = AMD64_NUM_COUNTERS, |
@@ -608,7 +760,7 @@ static __initconst const struct x86_pmu amd_pmu = { | |||
608 | 760 | ||
609 | static int setup_event_constraints(void) | 761 | static int setup_event_constraints(void) |
610 | { | 762 | { |
611 | if (boot_cpu_data.x86 >= 0x15) | 763 | if (boot_cpu_data.x86 == 0x15) |
612 | x86_pmu.get_event_constraints = amd_get_event_constraints_f15h; | 764 | x86_pmu.get_event_constraints = amd_get_event_constraints_f15h; |
613 | return 0; | 765 | return 0; |
614 | } | 766 | } |
@@ -638,6 +790,23 @@ static int setup_perfctr_core(void) | |||
638 | return 0; | 790 | return 0; |
639 | } | 791 | } |
640 | 792 | ||
793 | static int setup_perfctr_nb(void) | ||
794 | { | ||
795 | if (!cpu_has_perfctr_nb) | ||
796 | return -ENODEV; | ||
797 | |||
798 | x86_pmu.num_counters += AMD64_NUM_COUNTERS_NB; | ||
799 | |||
800 | if (cpu_has_perfctr_core) | ||
801 | amd_nb_event_constraint = &amd_NBPMC96; | ||
802 | else | ||
803 | amd_nb_event_constraint = &amd_NBPMC74; | ||
804 | |||
805 | printk(KERN_INFO "perf: AMD northbridge performance counters detected\n"); | ||
806 | |||
807 | return 0; | ||
808 | } | ||
809 | |||
641 | __init int amd_pmu_init(void) | 810 | __init int amd_pmu_init(void) |
642 | { | 811 | { |
643 | /* Performance-monitoring supported from K7 and later: */ | 812 | /* Performance-monitoring supported from K7 and later: */ |
@@ -648,6 +817,7 @@ __init int amd_pmu_init(void) | |||
648 | 817 | ||
649 | setup_event_constraints(); | 818 | setup_event_constraints(); |
650 | setup_perfctr_core(); | 819 | setup_perfctr_core(); |
820 | setup_perfctr_nb(); | ||
651 | 821 | ||
652 | /* Events are common for all AMDs */ | 822 | /* Events are common for all AMDs */ |
653 | memcpy(hw_cache_event_ids, amd_hw_cache_event_ids, | 823 | memcpy(hw_cache_event_ids, amd_hw_cache_event_ids, |
@@ -678,7 +848,7 @@ void amd_pmu_disable_virt(void) | |||
678 | * SVM is disabled the Guest-only bits still gets set and the counter | 848 | * SVM is disabled the Guest-only bits still gets set and the counter |
679 | * will not count anything. | 849 | * will not count anything. |
680 | */ | 850 | */ |
681 | cpuc->perf_ctr_virt_mask = AMD_PERFMON_EVENTSEL_HOSTONLY; | 851 | cpuc->perf_ctr_virt_mask = AMD64_EVENTSEL_HOSTONLY; |
682 | 852 | ||
683 | /* Reload all events */ | 853 | /* Reload all events */ |
684 | x86_pmu_disable_all(); | 854 | x86_pmu_disable_all(); |
diff --git a/arch/x86/kernel/kprobes/Makefile b/arch/x86/kernel/kprobes/Makefile new file mode 100644 index 000000000000..0d33169cc1a2 --- /dev/null +++ b/arch/x86/kernel/kprobes/Makefile | |||
@@ -0,0 +1,7 @@ | |||
1 | # | ||
2 | # Makefile for kernel probes | ||
3 | # | ||
4 | |||
5 | obj-$(CONFIG_KPROBES) += core.o | ||
6 | obj-$(CONFIG_OPTPROBES) += opt.o | ||
7 | obj-$(CONFIG_KPROBES_ON_FTRACE) += ftrace.o | ||
diff --git a/arch/x86/kernel/kprobes-common.h b/arch/x86/kernel/kprobes/common.h index 3230b68ef29a..2e9d4b5af036 100644 --- a/arch/x86/kernel/kprobes-common.h +++ b/arch/x86/kernel/kprobes/common.h | |||
@@ -99,4 +99,15 @@ static inline unsigned long __recover_optprobed_insn(kprobe_opcode_t *buf, unsig | |||
99 | return addr; | 99 | return addr; |
100 | } | 100 | } |
101 | #endif | 101 | #endif |
102 | |||
103 | #ifdef CONFIG_KPROBES_ON_FTRACE | ||
104 | extern int skip_singlestep(struct kprobe *p, struct pt_regs *regs, | ||
105 | struct kprobe_ctlblk *kcb); | ||
106 | #else | ||
107 | static inline int skip_singlestep(struct kprobe *p, struct pt_regs *regs, | ||
108 | struct kprobe_ctlblk *kcb) | ||
109 | { | ||
110 | return 0; | ||
111 | } | ||
112 | #endif | ||
102 | #endif | 113 | #endif |
diff --git a/arch/x86/kernel/kprobes.c b/arch/x86/kernel/kprobes/core.c index 57916c0d3cf6..e124554598ee 100644 --- a/arch/x86/kernel/kprobes.c +++ b/arch/x86/kernel/kprobes/core.c | |||
@@ -58,7 +58,7 @@ | |||
58 | #include <asm/insn.h> | 58 | #include <asm/insn.h> |
59 | #include <asm/debugreg.h> | 59 | #include <asm/debugreg.h> |
60 | 60 | ||
61 | #include "kprobes-common.h" | 61 | #include "common.h" |
62 | 62 | ||
63 | void jprobe_return_end(void); | 63 | void jprobe_return_end(void); |
64 | 64 | ||
@@ -78,7 +78,7 @@ DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk); | |||
78 | * Groups, and some special opcodes can not boost. | 78 | * Groups, and some special opcodes can not boost. |
79 | * This is non-const and volatile to keep gcc from statically | 79 | * This is non-const and volatile to keep gcc from statically |
80 | * optimizing it out, as variable_test_bit makes gcc think only | 80 | * optimizing it out, as variable_test_bit makes gcc think only |
81 | * *(unsigned long*) is used. | 81 | * *(unsigned long*) is used. |
82 | */ | 82 | */ |
83 | static volatile u32 twobyte_is_boostable[256 / 32] = { | 83 | static volatile u32 twobyte_is_boostable[256 / 32] = { |
84 | /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */ | 84 | /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */ |
@@ -117,7 +117,7 @@ static void __kprobes __synthesize_relative_insn(void *from, void *to, u8 op) | |||
117 | struct __arch_relative_insn { | 117 | struct __arch_relative_insn { |
118 | u8 op; | 118 | u8 op; |
119 | s32 raddr; | 119 | s32 raddr; |
120 | } __attribute__((packed)) *insn; | 120 | } __packed *insn; |
121 | 121 | ||
122 | insn = (struct __arch_relative_insn *)from; | 122 | insn = (struct __arch_relative_insn *)from; |
123 | insn->raddr = (s32)((long)(to) - ((long)(from) + 5)); | 123 | insn->raddr = (s32)((long)(to) - ((long)(from) + 5)); |
@@ -541,23 +541,6 @@ reenter_kprobe(struct kprobe *p, struct pt_regs *regs, struct kprobe_ctlblk *kcb | |||
541 | return 1; | 541 | return 1; |
542 | } | 542 | } |
543 | 543 | ||
544 | #ifdef KPROBES_CAN_USE_FTRACE | ||
545 | static void __kprobes skip_singlestep(struct kprobe *p, struct pt_regs *regs, | ||
546 | struct kprobe_ctlblk *kcb) | ||
547 | { | ||
548 | /* | ||
549 | * Emulate singlestep (and also recover regs->ip) | ||
550 | * as if there is a 5byte nop | ||
551 | */ | ||
552 | regs->ip = (unsigned long)p->addr + MCOUNT_INSN_SIZE; | ||
553 | if (unlikely(p->post_handler)) { | ||
554 | kcb->kprobe_status = KPROBE_HIT_SSDONE; | ||
555 | p->post_handler(p, regs, 0); | ||
556 | } | ||
557 | __this_cpu_write(current_kprobe, NULL); | ||
558 | } | ||
559 | #endif | ||
560 | |||
561 | /* | 544 | /* |
562 | * Interrupts are disabled on entry as trap3 is an interrupt gate and they | 545 | * Interrupts are disabled on entry as trap3 is an interrupt gate and they |
563 | * remain disabled throughout this function. | 546 | * remain disabled throughout this function. |
@@ -616,13 +599,8 @@ static int __kprobes kprobe_handler(struct pt_regs *regs) | |||
616 | } else if (kprobe_running()) { | 599 | } else if (kprobe_running()) { |
617 | p = __this_cpu_read(current_kprobe); | 600 | p = __this_cpu_read(current_kprobe); |
618 | if (p->break_handler && p->break_handler(p, regs)) { | 601 | if (p->break_handler && p->break_handler(p, regs)) { |
619 | #ifdef KPROBES_CAN_USE_FTRACE | 602 | if (!skip_singlestep(p, regs, kcb)) |
620 | if (kprobe_ftrace(p)) { | 603 | setup_singlestep(p, regs, kcb, 0); |
621 | skip_singlestep(p, regs, kcb); | ||
622 | return 1; | ||
623 | } | ||
624 | #endif | ||
625 | setup_singlestep(p, regs, kcb, 0); | ||
626 | return 1; | 604 | return 1; |
627 | } | 605 | } |
628 | } /* else: not a kprobe fault; let the kernel handle it */ | 606 | } /* else: not a kprobe fault; let the kernel handle it */ |
@@ -1075,50 +1053,6 @@ int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs) | |||
1075 | return 0; | 1053 | return 0; |
1076 | } | 1054 | } |
1077 | 1055 | ||
1078 | #ifdef KPROBES_CAN_USE_FTRACE | ||
1079 | /* Ftrace callback handler for kprobes */ | ||
1080 | void __kprobes kprobe_ftrace_handler(unsigned long ip, unsigned long parent_ip, | ||
1081 | struct ftrace_ops *ops, struct pt_regs *regs) | ||
1082 | { | ||
1083 | struct kprobe *p; | ||
1084 | struct kprobe_ctlblk *kcb; | ||
1085 | unsigned long flags; | ||
1086 | |||
1087 | /* Disable irq for emulating a breakpoint and avoiding preempt */ | ||
1088 | local_irq_save(flags); | ||
1089 | |||
1090 | p = get_kprobe((kprobe_opcode_t *)ip); | ||
1091 | if (unlikely(!p) || kprobe_disabled(p)) | ||
1092 | goto end; | ||
1093 | |||
1094 | kcb = get_kprobe_ctlblk(); | ||
1095 | if (kprobe_running()) { | ||
1096 | kprobes_inc_nmissed_count(p); | ||
1097 | } else { | ||
1098 | /* Kprobe handler expects regs->ip = ip + 1 as breakpoint hit */ | ||
1099 | regs->ip = ip + sizeof(kprobe_opcode_t); | ||
1100 | |||
1101 | __this_cpu_write(current_kprobe, p); | ||
1102 | kcb->kprobe_status = KPROBE_HIT_ACTIVE; | ||
1103 | if (!p->pre_handler || !p->pre_handler(p, regs)) | ||
1104 | skip_singlestep(p, regs, kcb); | ||
1105 | /* | ||
1106 | * If pre_handler returns !0, it sets regs->ip and | ||
1107 | * resets current kprobe. | ||
1108 | */ | ||
1109 | } | ||
1110 | end: | ||
1111 | local_irq_restore(flags); | ||
1112 | } | ||
1113 | |||
1114 | int __kprobes arch_prepare_kprobe_ftrace(struct kprobe *p) | ||
1115 | { | ||
1116 | p->ainsn.insn = NULL; | ||
1117 | p->ainsn.boostable = -1; | ||
1118 | return 0; | ||
1119 | } | ||
1120 | #endif | ||
1121 | |||
1122 | int __init arch_init_kprobes(void) | 1056 | int __init arch_init_kprobes(void) |
1123 | { | 1057 | { |
1124 | return arch_init_optprobes(); | 1058 | return arch_init_optprobes(); |
diff --git a/arch/x86/kernel/kprobes/ftrace.c b/arch/x86/kernel/kprobes/ftrace.c new file mode 100644 index 000000000000..23ef5c556f06 --- /dev/null +++ b/arch/x86/kernel/kprobes/ftrace.c | |||
@@ -0,0 +1,93 @@ | |||
1 | /* | ||
2 | * Dynamic Ftrace based Kprobes Optimization | ||
3 | * | ||
4 | * This program is free software; you can redistribute it and/or modify | ||
5 | * it under the terms of the GNU General Public License as published by | ||
6 | * the Free Software Foundation; either version 2 of the License, or | ||
7 | * (at your option) any later version. | ||
8 | * | ||
9 | * This program is distributed in the hope that it will be useful, | ||
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
12 | * GNU General Public License for more details. | ||
13 | * | ||
14 | * You should have received a copy of the GNU General Public License | ||
15 | * along with this program; if not, write to the Free Software | ||
16 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | ||
17 | * | ||
18 | * Copyright (C) Hitachi Ltd., 2012 | ||
19 | */ | ||
20 | #include <linux/kprobes.h> | ||
21 | #include <linux/ptrace.h> | ||
22 | #include <linux/hardirq.h> | ||
23 | #include <linux/preempt.h> | ||
24 | #include <linux/ftrace.h> | ||
25 | |||
26 | #include "common.h" | ||
27 | |||
28 | static int __skip_singlestep(struct kprobe *p, struct pt_regs *regs, | ||
29 | struct kprobe_ctlblk *kcb) | ||
30 | { | ||
31 | /* | ||
32 | * Emulate singlestep (and also recover regs->ip) | ||
33 | * as if there is a 5byte nop | ||
34 | */ | ||
35 | regs->ip = (unsigned long)p->addr + MCOUNT_INSN_SIZE; | ||
36 | if (unlikely(p->post_handler)) { | ||
37 | kcb->kprobe_status = KPROBE_HIT_SSDONE; | ||
38 | p->post_handler(p, regs, 0); | ||
39 | } | ||
40 | __this_cpu_write(current_kprobe, NULL); | ||
41 | return 1; | ||
42 | } | ||
43 | |||
44 | int __kprobes skip_singlestep(struct kprobe *p, struct pt_regs *regs, | ||
45 | struct kprobe_ctlblk *kcb) | ||
46 | { | ||
47 | if (kprobe_ftrace(p)) | ||
48 | return __skip_singlestep(p, regs, kcb); | ||
49 | else | ||
50 | return 0; | ||
51 | } | ||
52 | |||
53 | /* Ftrace callback handler for kprobes */ | ||
54 | void __kprobes kprobe_ftrace_handler(unsigned long ip, unsigned long parent_ip, | ||
55 | struct ftrace_ops *ops, struct pt_regs *regs) | ||
56 | { | ||
57 | struct kprobe *p; | ||
58 | struct kprobe_ctlblk *kcb; | ||
59 | unsigned long flags; | ||
60 | |||
61 | /* Disable irq for emulating a breakpoint and avoiding preempt */ | ||
62 | local_irq_save(flags); | ||
63 | |||
64 | p = get_kprobe((kprobe_opcode_t *)ip); | ||
65 | if (unlikely(!p) || kprobe_disabled(p)) | ||
66 | goto end; | ||
67 | |||
68 | kcb = get_kprobe_ctlblk(); | ||
69 | if (kprobe_running()) { | ||
70 | kprobes_inc_nmissed_count(p); | ||
71 | } else { | ||
72 | /* Kprobe handler expects regs->ip = ip + 1 as breakpoint hit */ | ||
73 | regs->ip = ip + sizeof(kprobe_opcode_t); | ||
74 | |||
75 | __this_cpu_write(current_kprobe, p); | ||
76 | kcb->kprobe_status = KPROBE_HIT_ACTIVE; | ||
77 | if (!p->pre_handler || !p->pre_handler(p, regs)) | ||
78 | __skip_singlestep(p, regs, kcb); | ||
79 | /* | ||
80 | * If pre_handler returns !0, it sets regs->ip and | ||
81 | * resets current kprobe. | ||
82 | */ | ||
83 | } | ||
84 | end: | ||
85 | local_irq_restore(flags); | ||
86 | } | ||
87 | |||
88 | int __kprobes arch_prepare_kprobe_ftrace(struct kprobe *p) | ||
89 | { | ||
90 | p->ainsn.insn = NULL; | ||
91 | p->ainsn.boostable = -1; | ||
92 | return 0; | ||
93 | } | ||
diff --git a/arch/x86/kernel/kprobes-opt.c b/arch/x86/kernel/kprobes/opt.c index c5e410eed403..76dc6f095724 100644 --- a/arch/x86/kernel/kprobes-opt.c +++ b/arch/x86/kernel/kprobes/opt.c | |||
@@ -37,7 +37,7 @@ | |||
37 | #include <asm/insn.h> | 37 | #include <asm/insn.h> |
38 | #include <asm/debugreg.h> | 38 | #include <asm/debugreg.h> |
39 | 39 | ||
40 | #include "kprobes-common.h" | 40 | #include "common.h" |
41 | 41 | ||
42 | unsigned long __recover_optprobed_insn(kprobe_opcode_t *buf, unsigned long addr) | 42 | unsigned long __recover_optprobed_insn(kprobe_opcode_t *buf, unsigned long addr) |
43 | { | 43 | { |
diff --git a/arch/x86/kernel/uprobes.c b/arch/x86/kernel/uprobes.c index c71025b67462..0ba4cfb4f412 100644 --- a/arch/x86/kernel/uprobes.c +++ b/arch/x86/kernel/uprobes.c | |||
@@ -680,8 +680,10 @@ static bool __skip_sstep(struct arch_uprobe *auprobe, struct pt_regs *regs) | |||
680 | if (auprobe->insn[i] == 0x66) | 680 | if (auprobe->insn[i] == 0x66) |
681 | continue; | 681 | continue; |
682 | 682 | ||
683 | if (auprobe->insn[i] == 0x90) | 683 | if (auprobe->insn[i] == 0x90) { |
684 | regs->ip += i + 1; | ||
684 | return true; | 685 | return true; |
686 | } | ||
685 | 687 | ||
686 | break; | 688 | break; |
687 | } | 689 | } |