diff options
Diffstat (limited to 'kernel')
-rw-r--r-- | kernel/smp.c | 18 | ||||
-rw-r--r-- | kernel/timer.c | 129 |
2 files changed, 111 insertions, 36 deletions
diff --git a/kernel/smp.c b/kernel/smp.c index f362a8553777..75c8dde58c55 100644 --- a/kernel/smp.c +++ b/kernel/smp.c | |||
@@ -51,10 +51,6 @@ static void csd_flag_wait(struct call_single_data *data) | |||
51 | { | 51 | { |
52 | /* Wait for response */ | 52 | /* Wait for response */ |
53 | do { | 53 | do { |
54 | /* | ||
55 | * We need to see the flags store in the IPI handler | ||
56 | */ | ||
57 | smp_mb(); | ||
58 | if (!(data->flags & CSD_FLAG_WAIT)) | 54 | if (!(data->flags & CSD_FLAG_WAIT)) |
59 | break; | 55 | break; |
60 | cpu_relax(); | 56 | cpu_relax(); |
@@ -76,6 +72,11 @@ static void generic_exec_single(int cpu, struct call_single_data *data) | |||
76 | list_add_tail(&data->list, &dst->list); | 72 | list_add_tail(&data->list, &dst->list); |
77 | spin_unlock_irqrestore(&dst->lock, flags); | 73 | spin_unlock_irqrestore(&dst->lock, flags); |
78 | 74 | ||
75 | /* | ||
76 | * Make the list addition visible before sending the ipi. | ||
77 | */ | ||
78 | smp_mb(); | ||
79 | |||
79 | if (ipi) | 80 | if (ipi) |
80 | arch_send_call_function_single_ipi(cpu); | 81 | arch_send_call_function_single_ipi(cpu); |
81 | 82 | ||
@@ -157,7 +158,7 @@ void generic_smp_call_function_single_interrupt(void) | |||
157 | * Need to see other stores to list head for checking whether | 158 | * Need to see other stores to list head for checking whether |
158 | * list is empty without holding q->lock | 159 | * list is empty without holding q->lock |
159 | */ | 160 | */ |
160 | smp_mb(); | 161 | smp_read_barrier_depends(); |
161 | while (!list_empty(&q->list)) { | 162 | while (!list_empty(&q->list)) { |
162 | unsigned int data_flags; | 163 | unsigned int data_flags; |
163 | 164 | ||
@@ -191,7 +192,7 @@ void generic_smp_call_function_single_interrupt(void) | |||
191 | /* | 192 | /* |
192 | * See comment on outer loop | 193 | * See comment on outer loop |
193 | */ | 194 | */ |
194 | smp_mb(); | 195 | smp_read_barrier_depends(); |
195 | } | 196 | } |
196 | } | 197 | } |
197 | 198 | ||
@@ -370,6 +371,11 @@ int smp_call_function_mask(cpumask_t mask, void (*func)(void *), void *info, | |||
370 | list_add_tail_rcu(&data->csd.list, &call_function_queue); | 371 | list_add_tail_rcu(&data->csd.list, &call_function_queue); |
371 | spin_unlock_irqrestore(&call_function_lock, flags); | 372 | spin_unlock_irqrestore(&call_function_lock, flags); |
372 | 373 | ||
374 | /* | ||
375 | * Make the list addition visible before sending the ipi. | ||
376 | */ | ||
377 | smp_mb(); | ||
378 | |||
373 | /* Send a message to all CPUs in the map */ | 379 | /* Send a message to all CPUs in the map */ |
374 | arch_send_call_function_ipi(mask); | 380 | arch_send_call_function_ipi(mask); |
375 | 381 | ||
diff --git a/kernel/timer.c b/kernel/timer.c index 56becf373c58..dbd50fabe4c7 100644 --- a/kernel/timer.c +++ b/kernel/timer.c | |||
@@ -112,27 +112,8 @@ timer_set_base(struct timer_list *timer, struct tvec_base *new_base) | |||
112 | tbase_get_deferrable(timer->base)); | 112 | tbase_get_deferrable(timer->base)); |
113 | } | 113 | } |
114 | 114 | ||
115 | /** | 115 | static unsigned long round_jiffies_common(unsigned long j, int cpu, |
116 | * __round_jiffies - function to round jiffies to a full second | 116 | bool force_up) |
117 | * @j: the time in (absolute) jiffies that should be rounded | ||
118 | * @cpu: the processor number on which the timeout will happen | ||
119 | * | ||
120 | * __round_jiffies() rounds an absolute time in the future (in jiffies) | ||
121 | * up or down to (approximately) full seconds. This is useful for timers | ||
122 | * for which the exact time they fire does not matter too much, as long as | ||
123 | * they fire approximately every X seconds. | ||
124 | * | ||
125 | * By rounding these timers to whole seconds, all such timers will fire | ||
126 | * at the same time, rather than at various times spread out. The goal | ||
127 | * of this is to have the CPU wake up less, which saves power. | ||
128 | * | ||
129 | * The exact rounding is skewed for each processor to avoid all | ||
130 | * processors firing at the exact same time, which could lead | ||
131 | * to lock contention or spurious cache line bouncing. | ||
132 | * | ||
133 | * The return value is the rounded version of the @j parameter. | ||
134 | */ | ||
135 | unsigned long __round_jiffies(unsigned long j, int cpu) | ||
136 | { | 117 | { |
137 | int rem; | 118 | int rem; |
138 | unsigned long original = j; | 119 | unsigned long original = j; |
@@ -154,8 +135,9 @@ unsigned long __round_jiffies(unsigned long j, int cpu) | |||
154 | * due to delays of the timer irq, long irq off times etc etc) then | 135 | * due to delays of the timer irq, long irq off times etc etc) then |
155 | * we should round down to the whole second, not up. Use 1/4th second | 136 | * we should round down to the whole second, not up. Use 1/4th second |
156 | * as cutoff for this rounding as an extreme upper bound for this. | 137 | * as cutoff for this rounding as an extreme upper bound for this. |
138 | * But never round down if @force_up is set. | ||
157 | */ | 139 | */ |
158 | if (rem < HZ/4) /* round down */ | 140 | if (rem < HZ/4 && !force_up) /* round down */ |
159 | j = j - rem; | 141 | j = j - rem; |
160 | else /* round up */ | 142 | else /* round up */ |
161 | j = j - rem + HZ; | 143 | j = j - rem + HZ; |
@@ -167,6 +149,31 @@ unsigned long __round_jiffies(unsigned long j, int cpu) | |||
167 | return original; | 149 | return original; |
168 | return j; | 150 | return j; |
169 | } | 151 | } |
152 | |||
153 | /** | ||
154 | * __round_jiffies - function to round jiffies to a full second | ||
155 | * @j: the time in (absolute) jiffies that should be rounded | ||
156 | * @cpu: the processor number on which the timeout will happen | ||
157 | * | ||
158 | * __round_jiffies() rounds an absolute time in the future (in jiffies) | ||
159 | * up or down to (approximately) full seconds. This is useful for timers | ||
160 | * for which the exact time they fire does not matter too much, as long as | ||
161 | * they fire approximately every X seconds. | ||
162 | * | ||
163 | * By rounding these timers to whole seconds, all such timers will fire | ||
164 | * at the same time, rather than at various times spread out. The goal | ||
165 | * of this is to have the CPU wake up less, which saves power. | ||
166 | * | ||
167 | * The exact rounding is skewed for each processor to avoid all | ||
168 | * processors firing at the exact same time, which could lead | ||
169 | * to lock contention or spurious cache line bouncing. | ||
170 | * | ||
171 | * The return value is the rounded version of the @j parameter. | ||
172 | */ | ||
173 | unsigned long __round_jiffies(unsigned long j, int cpu) | ||
174 | { | ||
175 | return round_jiffies_common(j, cpu, false); | ||
176 | } | ||
170 | EXPORT_SYMBOL_GPL(__round_jiffies); | 177 | EXPORT_SYMBOL_GPL(__round_jiffies); |
171 | 178 | ||
172 | /** | 179 | /** |
@@ -191,13 +198,10 @@ EXPORT_SYMBOL_GPL(__round_jiffies); | |||
191 | */ | 198 | */ |
192 | unsigned long __round_jiffies_relative(unsigned long j, int cpu) | 199 | unsigned long __round_jiffies_relative(unsigned long j, int cpu) |
193 | { | 200 | { |
194 | /* | 201 | unsigned long j0 = jiffies; |
195 | * In theory the following code can skip a jiffy in case jiffies | 202 | |
196 | * increments right between the addition and the later subtraction. | 203 | /* Use j0 because jiffies might change while we run */ |
197 | * However since the entire point of this function is to use approximate | 204 | return round_jiffies_common(j + j0, cpu, false) - j0; |
198 | * timeouts, it's entirely ok to not handle that. | ||
199 | */ | ||
200 | return __round_jiffies(j + jiffies, cpu) - jiffies; | ||
201 | } | 205 | } |
202 | EXPORT_SYMBOL_GPL(__round_jiffies_relative); | 206 | EXPORT_SYMBOL_GPL(__round_jiffies_relative); |
203 | 207 | ||
@@ -218,7 +222,7 @@ EXPORT_SYMBOL_GPL(__round_jiffies_relative); | |||
218 | */ | 222 | */ |
219 | unsigned long round_jiffies(unsigned long j) | 223 | unsigned long round_jiffies(unsigned long j) |
220 | { | 224 | { |
221 | return __round_jiffies(j, raw_smp_processor_id()); | 225 | return round_jiffies_common(j, raw_smp_processor_id(), false); |
222 | } | 226 | } |
223 | EXPORT_SYMBOL_GPL(round_jiffies); | 227 | EXPORT_SYMBOL_GPL(round_jiffies); |
224 | 228 | ||
@@ -243,6 +247,71 @@ unsigned long round_jiffies_relative(unsigned long j) | |||
243 | } | 247 | } |
244 | EXPORT_SYMBOL_GPL(round_jiffies_relative); | 248 | EXPORT_SYMBOL_GPL(round_jiffies_relative); |
245 | 249 | ||
250 | /** | ||
251 | * __round_jiffies_up - function to round jiffies up to a full second | ||
252 | * @j: the time in (absolute) jiffies that should be rounded | ||
253 | * @cpu: the processor number on which the timeout will happen | ||
254 | * | ||
255 | * This is the same as __round_jiffies() except that it will never | ||
256 | * round down. This is useful for timeouts for which the exact time | ||
257 | * of firing does not matter too much, as long as they don't fire too | ||
258 | * early. | ||
259 | */ | ||
260 | unsigned long __round_jiffies_up(unsigned long j, int cpu) | ||
261 | { | ||
262 | return round_jiffies_common(j, cpu, true); | ||
263 | } | ||
264 | EXPORT_SYMBOL_GPL(__round_jiffies_up); | ||
265 | |||
266 | /** | ||
267 | * __round_jiffies_up_relative - function to round jiffies up to a full second | ||
268 | * @j: the time in (relative) jiffies that should be rounded | ||
269 | * @cpu: the processor number on which the timeout will happen | ||
270 | * | ||
271 | * This is the same as __round_jiffies_relative() except that it will never | ||
272 | * round down. This is useful for timeouts for which the exact time | ||
273 | * of firing does not matter too much, as long as they don't fire too | ||
274 | * early. | ||
275 | */ | ||
276 | unsigned long __round_jiffies_up_relative(unsigned long j, int cpu) | ||
277 | { | ||
278 | unsigned long j0 = jiffies; | ||
279 | |||
280 | /* Use j0 because jiffies might change while we run */ | ||
281 | return round_jiffies_common(j + j0, cpu, true) - j0; | ||
282 | } | ||
283 | EXPORT_SYMBOL_GPL(__round_jiffies_up_relative); | ||
284 | |||
285 | /** | ||
286 | * round_jiffies_up - function to round jiffies up to a full second | ||
287 | * @j: the time in (absolute) jiffies that should be rounded | ||
288 | * | ||
289 | * This is the same as round_jiffies() except that it will never | ||
290 | * round down. This is useful for timeouts for which the exact time | ||
291 | * of firing does not matter too much, as long as they don't fire too | ||
292 | * early. | ||
293 | */ | ||
294 | unsigned long round_jiffies_up(unsigned long j) | ||
295 | { | ||
296 | return round_jiffies_common(j, raw_smp_processor_id(), true); | ||
297 | } | ||
298 | EXPORT_SYMBOL_GPL(round_jiffies_up); | ||
299 | |||
300 | /** | ||
301 | * round_jiffies_up_relative - function to round jiffies up to a full second | ||
302 | * @j: the time in (relative) jiffies that should be rounded | ||
303 | * | ||
304 | * This is the same as round_jiffies_relative() except that it will never | ||
305 | * round down. This is useful for timeouts for which the exact time | ||
306 | * of firing does not matter too much, as long as they don't fire too | ||
307 | * early. | ||
308 | */ | ||
309 | unsigned long round_jiffies_up_relative(unsigned long j) | ||
310 | { | ||
311 | return __round_jiffies_up_relative(j, raw_smp_processor_id()); | ||
312 | } | ||
313 | EXPORT_SYMBOL_GPL(round_jiffies_up_relative); | ||
314 | |||
246 | 315 | ||
247 | static inline void set_running_timer(struct tvec_base *base, | 316 | static inline void set_running_timer(struct tvec_base *base, |
248 | struct timer_list *timer) | 317 | struct timer_list *timer) |