diff options
Diffstat (limited to 'drivers/oprofile/cpu_buffer.c')
-rw-r--r-- | drivers/oprofile/cpu_buffer.c | 75 |
1 files changed, 23 insertions, 52 deletions
diff --git a/drivers/oprofile/cpu_buffer.c b/drivers/oprofile/cpu_buffer.c index 166b67ea622f..219f79e2210a 100644 --- a/drivers/oprofile/cpu_buffer.c +++ b/drivers/oprofile/cpu_buffer.c | |||
@@ -30,23 +30,7 @@ | |||
30 | 30 | ||
31 | #define OP_BUFFER_FLAGS 0 | 31 | #define OP_BUFFER_FLAGS 0 |
32 | 32 | ||
33 | /* | 33 | static struct ring_buffer *op_ring_buffer; |
34 | * Read and write access is using spin locking. Thus, writing to the | ||
35 | * buffer by NMI handler (x86) could occur also during critical | ||
36 | * sections when reading the buffer. To avoid this, there are 2 | ||
37 | * buffers for independent read and write access. Read access is in | ||
38 | * process context only, write access only in the NMI handler. If the | ||
39 | * read buffer runs empty, both buffers are swapped atomically. There | ||
40 | * is potentially a small window during swapping where the buffers are | ||
41 | * disabled and samples could be lost. | ||
42 | * | ||
43 | * Using 2 buffers is a little bit overhead, but the solution is clear | ||
44 | * and does not require changes in the ring buffer implementation. It | ||
45 | * can be changed to a single buffer solution when the ring buffer | ||
46 | * access is implemented as non-locking atomic code. | ||
47 | */ | ||
48 | static struct ring_buffer *op_ring_buffer_read; | ||
49 | static struct ring_buffer *op_ring_buffer_write; | ||
50 | DEFINE_PER_CPU(struct oprofile_cpu_buffer, op_cpu_buffer); | 34 | DEFINE_PER_CPU(struct oprofile_cpu_buffer, op_cpu_buffer); |
51 | 35 | ||
52 | static void wq_sync_buffer(struct work_struct *work); | 36 | static void wq_sync_buffer(struct work_struct *work); |
@@ -68,12 +52,9 @@ void oprofile_cpu_buffer_inc_smpl_lost(void) | |||
68 | 52 | ||
69 | void free_cpu_buffers(void) | 53 | void free_cpu_buffers(void) |
70 | { | 54 | { |
71 | if (op_ring_buffer_read) | 55 | if (op_ring_buffer) |
72 | ring_buffer_free(op_ring_buffer_read); | 56 | ring_buffer_free(op_ring_buffer); |
73 | op_ring_buffer_read = NULL; | 57 | op_ring_buffer = NULL; |
74 | if (op_ring_buffer_write) | ||
75 | ring_buffer_free(op_ring_buffer_write); | ||
76 | op_ring_buffer_write = NULL; | ||
77 | } | 58 | } |
78 | 59 | ||
79 | #define RB_EVENT_HDR_SIZE 4 | 60 | #define RB_EVENT_HDR_SIZE 4 |
@@ -86,11 +67,8 @@ int alloc_cpu_buffers(void) | |||
86 | unsigned long byte_size = buffer_size * (sizeof(struct op_sample) + | 67 | unsigned long byte_size = buffer_size * (sizeof(struct op_sample) + |
87 | RB_EVENT_HDR_SIZE); | 68 | RB_EVENT_HDR_SIZE); |
88 | 69 | ||
89 | op_ring_buffer_read = ring_buffer_alloc(byte_size, OP_BUFFER_FLAGS); | 70 | op_ring_buffer = ring_buffer_alloc(byte_size, OP_BUFFER_FLAGS); |
90 | if (!op_ring_buffer_read) | 71 | if (!op_ring_buffer) |
91 | goto fail; | ||
92 | op_ring_buffer_write = ring_buffer_alloc(byte_size, OP_BUFFER_FLAGS); | ||
93 | if (!op_ring_buffer_write) | ||
94 | goto fail; | 72 | goto fail; |
95 | 73 | ||
96 | for_each_possible_cpu(i) { | 74 | for_each_possible_cpu(i) { |
@@ -162,16 +140,11 @@ struct op_sample | |||
162 | *op_cpu_buffer_write_reserve(struct op_entry *entry, unsigned long size) | 140 | *op_cpu_buffer_write_reserve(struct op_entry *entry, unsigned long size) |
163 | { | 141 | { |
164 | entry->event = ring_buffer_lock_reserve | 142 | entry->event = ring_buffer_lock_reserve |
165 | (op_ring_buffer_write, sizeof(struct op_sample) + | 143 | (op_ring_buffer, sizeof(struct op_sample) + |
166 | size * sizeof(entry->sample->data[0])); | 144 | size * sizeof(entry->sample->data[0])); |
167 | if (entry->event) | 145 | if (!entry->event) |
168 | entry->sample = ring_buffer_event_data(entry->event); | ||
169 | else | ||
170 | entry->sample = NULL; | ||
171 | |||
172 | if (!entry->sample) | ||
173 | return NULL; | 146 | return NULL; |
174 | 147 | entry->sample = ring_buffer_event_data(entry->event); | |
175 | entry->size = size; | 148 | entry->size = size; |
176 | entry->data = entry->sample->data; | 149 | entry->data = entry->sample->data; |
177 | 150 | ||
@@ -180,25 +153,16 @@ struct op_sample | |||
180 | 153 | ||
181 | int op_cpu_buffer_write_commit(struct op_entry *entry) | 154 | int op_cpu_buffer_write_commit(struct op_entry *entry) |
182 | { | 155 | { |
183 | return ring_buffer_unlock_commit(op_ring_buffer_write, entry->event); | 156 | return ring_buffer_unlock_commit(op_ring_buffer, entry->event); |
184 | } | 157 | } |
185 | 158 | ||
186 | struct op_sample *op_cpu_buffer_read_entry(struct op_entry *entry, int cpu) | 159 | struct op_sample *op_cpu_buffer_read_entry(struct op_entry *entry, int cpu) |
187 | { | 160 | { |
188 | struct ring_buffer_event *e; | 161 | struct ring_buffer_event *e; |
189 | e = ring_buffer_consume(op_ring_buffer_read, cpu, NULL); | 162 | e = ring_buffer_consume(op_ring_buffer, cpu, NULL, NULL); |
190 | if (e) | 163 | if (!e) |
191 | goto event; | ||
192 | if (ring_buffer_swap_cpu(op_ring_buffer_read, | ||
193 | op_ring_buffer_write, | ||
194 | cpu)) | ||
195 | return NULL; | 164 | return NULL; |
196 | e = ring_buffer_consume(op_ring_buffer_read, cpu, NULL); | ||
197 | if (e) | ||
198 | goto event; | ||
199 | return NULL; | ||
200 | 165 | ||
201 | event: | ||
202 | entry->event = e; | 166 | entry->event = e; |
203 | entry->sample = ring_buffer_event_data(e); | 167 | entry->sample = ring_buffer_event_data(e); |
204 | entry->size = (ring_buffer_event_length(e) - sizeof(struct op_sample)) | 168 | entry->size = (ring_buffer_event_length(e) - sizeof(struct op_sample)) |
@@ -209,8 +173,7 @@ event: | |||
209 | 173 | ||
210 | unsigned long op_cpu_buffer_entries(int cpu) | 174 | unsigned long op_cpu_buffer_entries(int cpu) |
211 | { | 175 | { |
212 | return ring_buffer_entries_cpu(op_ring_buffer_read, cpu) | 176 | return ring_buffer_entries_cpu(op_ring_buffer, cpu); |
213 | + ring_buffer_entries_cpu(op_ring_buffer_write, cpu); | ||
214 | } | 177 | } |
215 | 178 | ||
216 | static int | 179 | static int |
@@ -356,8 +319,16 @@ void oprofile_add_ext_sample(unsigned long pc, struct pt_regs * const regs, | |||
356 | 319 | ||
357 | void oprofile_add_sample(struct pt_regs * const regs, unsigned long event) | 320 | void oprofile_add_sample(struct pt_regs * const regs, unsigned long event) |
358 | { | 321 | { |
359 | int is_kernel = !user_mode(regs); | 322 | int is_kernel; |
360 | unsigned long pc = profile_pc(regs); | 323 | unsigned long pc; |
324 | |||
325 | if (likely(regs)) { | ||
326 | is_kernel = !user_mode(regs); | ||
327 | pc = profile_pc(regs); | ||
328 | } else { | ||
329 | is_kernel = 0; /* This value will not be used */ | ||
330 | pc = ESCAPE_CODE; /* as this causes an early return. */ | ||
331 | } | ||
361 | 332 | ||
362 | __oprofile_add_ext_sample(pc, regs, event, is_kernel); | 333 | __oprofile_add_ext_sample(pc, regs, event, is_kernel); |
363 | } | 334 | } |