Commit | Line | Data |
---|---|---|
f6ac2354 CL |
1 | /* |
2 | * linux/mm/vmstat.c | |
3 | * | |
4 | * Manages VM statistics | |
5 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
2244b95a CL |
6 | * |
7 | * zoned VM statistics | |
8 | * Copyright (C) 2006 Silicon Graphics, Inc., | |
9 | * Christoph Lameter <christoph@lameter.com> | |
f6ac2354 | 10 | */ |
8f32f7e5 | 11 | #include <linux/fs.h> |
f6ac2354 | 12 | #include <linux/mm.h> |
4e950f6f | 13 | #include <linux/err.h> |
2244b95a | 14 | #include <linux/module.h> |
5a0e3ad6 | 15 | #include <linux/slab.h> |
df9ecaba | 16 | #include <linux/cpu.h> |
c748e134 | 17 | #include <linux/vmstat.h> |
e8edc6e0 | 18 | #include <linux/sched.h> |
f1a5ab12 | 19 | #include <linux/math64.h> |
79da826a | 20 | #include <linux/writeback.h> |
36deb0be | 21 | #include <linux/compaction.h> |
6e543d57 LD |
22 | #include <linux/mm_inline.h> |
23 | ||
24 | #include "internal.h" | |
f6ac2354 | 25 | |
f8891e5e CL |
26 | #ifdef CONFIG_VM_EVENT_COUNTERS |
27 | DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}}; | |
28 | EXPORT_PER_CPU_SYMBOL(vm_event_states); | |
29 | ||
31f961a8 | 30 | static void sum_vm_events(unsigned long *ret) |
f8891e5e | 31 | { |
9eccf2a8 | 32 | int cpu; |
f8891e5e CL |
33 | int i; |
34 | ||
35 | memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long)); | |
36 | ||
31f961a8 | 37 | for_each_online_cpu(cpu) { |
f8891e5e CL |
38 | struct vm_event_state *this = &per_cpu(vm_event_states, cpu); |
39 | ||
f8891e5e CL |
40 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) |
41 | ret[i] += this->event[i]; | |
42 | } | |
43 | } | |
44 | ||
45 | /* | |
46 | * Accumulate the vm event counters across all CPUs. | |
47 | * The result is unavoidably approximate - it can change | |
48 | * during and after execution of this function. | |
49 | */ | |
50 | void all_vm_events(unsigned long *ret) | |
51 | { | |
b5be1132 | 52 | get_online_cpus(); |
31f961a8 | 53 | sum_vm_events(ret); |
b5be1132 | 54 | put_online_cpus(); |
f8891e5e | 55 | } |
32dd66fc | 56 | EXPORT_SYMBOL_GPL(all_vm_events); |
f8891e5e | 57 | |
f8891e5e CL |
58 | /* |
59 | * Fold the foreign cpu events into our own. | |
60 | * | |
61 | * This is adding to the events on one processor | |
62 | * but keeps the global counts constant. | |
63 | */ | |
64 | void vm_events_fold_cpu(int cpu) | |
65 | { | |
66 | struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu); | |
67 | int i; | |
68 | ||
69 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { | |
70 | count_vm_events(i, fold_state->event[i]); | |
71 | fold_state->event[i] = 0; | |
72 | } | |
73 | } | |
f8891e5e CL |
74 | |
75 | #endif /* CONFIG_VM_EVENT_COUNTERS */ | |
76 | ||
2244b95a CL |
77 | /* |
78 | * Manage combined zone based / global counters | |
79 | * | |
80 | * vm_stat contains the global counters | |
81 | */ | |
a1cb2c60 | 82 | atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS] __cacheline_aligned_in_smp; |
2244b95a CL |
83 | EXPORT_SYMBOL(vm_stat); |
84 | ||
85 | #ifdef CONFIG_SMP | |
86 | ||
b44129b3 | 87 | int calculate_pressure_threshold(struct zone *zone) |
88f5acf8 MG |
88 | { |
89 | int threshold; | |
90 | int watermark_distance; | |
91 | ||
92 | /* | |
93 | * As vmstats are not up to date, there is drift between the estimated | |
94 | * and real values. For high thresholds and a high number of CPUs, it | |
95 | * is possible for the min watermark to be breached while the estimated | |
96 | * value looks fine. The pressure threshold is a reduced value such | |
97 | * that even the maximum amount of drift will not accidentally breach | |
98 | * the min watermark | |
99 | */ | |
100 | watermark_distance = low_wmark_pages(zone) - min_wmark_pages(zone); | |
101 | threshold = max(1, (int)(watermark_distance / num_online_cpus())); | |
102 | ||
103 | /* | |
104 | * Maximum threshold is 125 | |
105 | */ | |
106 | threshold = min(125, threshold); | |
107 | ||
108 | return threshold; | |
109 | } | |
110 | ||
b44129b3 | 111 | int calculate_normal_threshold(struct zone *zone) |
df9ecaba CL |
112 | { |
113 | int threshold; | |
114 | int mem; /* memory in 128 MB units */ | |
115 | ||
116 | /* | |
117 | * The threshold scales with the number of processors and the amount | |
118 | * of memory per zone. More memory means that we can defer updates for | |
119 | * longer, more processors could lead to more contention. | |
120 | * fls() is used to have a cheap way of logarithmic scaling. | |
121 | * | |
122 | * Some sample thresholds: | |
123 | * | |
124 | * Threshold Processors (fls) Zonesize fls(mem+1) | |
125 | * ------------------------------------------------------------------ | |
126 | * 8 1 1 0.9-1 GB 4 | |
127 | * 16 2 2 0.9-1 GB 4 | |
128 | * 20 2 2 1-2 GB 5 | |
129 | * 24 2 2 2-4 GB 6 | |
130 | * 28 2 2 4-8 GB 7 | |
131 | * 32 2 2 8-16 GB 8 | |
132 | * 4 2 2 <128M 1 | |
133 | * 30 4 3 2-4 GB 5 | |
134 | * 48 4 3 8-16 GB 8 | |
135 | * 32 8 4 1-2 GB 4 | |
136 | * 32 8 4 0.9-1GB 4 | |
137 | * 10 16 5 <128M 1 | |
138 | * 40 16 5 900M 4 | |
139 | * 70 64 7 2-4 GB 5 | |
140 | * 84 64 7 4-8 GB 6 | |
141 | * 108 512 9 4-8 GB 6 | |
142 | * 125 1024 10 8-16 GB 8 | |
143 | * 125 1024 10 16-32 GB 9 | |
144 | */ | |
145 | ||
b40da049 | 146 | mem = zone->managed_pages >> (27 - PAGE_SHIFT); |
df9ecaba CL |
147 | |
148 | threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem)); | |
149 | ||
150 | /* | |
151 | * Maximum threshold is 125 | |
152 | */ | |
153 | threshold = min(125, threshold); | |
154 | ||
155 | return threshold; | |
156 | } | |
2244b95a CL |
157 | |
158 | /* | |
df9ecaba | 159 | * Refresh the thresholds for each zone. |
2244b95a | 160 | */ |
a6cccdc3 | 161 | void refresh_zone_stat_thresholds(void) |
2244b95a | 162 | { |
df9ecaba CL |
163 | struct zone *zone; |
164 | int cpu; | |
165 | int threshold; | |
166 | ||
ee99c71c | 167 | for_each_populated_zone(zone) { |
aa454840 CL |
168 | unsigned long max_drift, tolerate_drift; |
169 | ||
b44129b3 | 170 | threshold = calculate_normal_threshold(zone); |
df9ecaba CL |
171 | |
172 | for_each_online_cpu(cpu) | |
99dcc3e5 CL |
173 | per_cpu_ptr(zone->pageset, cpu)->stat_threshold |
174 | = threshold; | |
aa454840 CL |
175 | |
176 | /* | |
177 | * Only set percpu_drift_mark if there is a danger that | |
178 | * NR_FREE_PAGES reports the low watermark is ok when in fact | |
179 | * the min watermark could be breached by an allocation | |
180 | */ | |
181 | tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone); | |
182 | max_drift = num_online_cpus() * threshold; | |
183 | if (max_drift > tolerate_drift) | |
184 | zone->percpu_drift_mark = high_wmark_pages(zone) + | |
185 | max_drift; | |
df9ecaba | 186 | } |
2244b95a CL |
187 | } |
188 | ||
b44129b3 MG |
189 | void set_pgdat_percpu_threshold(pg_data_t *pgdat, |
190 | int (*calculate_pressure)(struct zone *)) | |
88f5acf8 MG |
191 | { |
192 | struct zone *zone; | |
193 | int cpu; | |
194 | int threshold; | |
195 | int i; | |
196 | ||
88f5acf8 MG |
197 | for (i = 0; i < pgdat->nr_zones; i++) { |
198 | zone = &pgdat->node_zones[i]; | |
199 | if (!zone->percpu_drift_mark) | |
200 | continue; | |
201 | ||
b44129b3 MG |
202 | threshold = (*calculate_pressure)(zone); |
203 | for_each_possible_cpu(cpu) | |
88f5acf8 MG |
204 | per_cpu_ptr(zone->pageset, cpu)->stat_threshold |
205 | = threshold; | |
206 | } | |
88f5acf8 MG |
207 | } |
208 | ||
2244b95a | 209 | /* |
bea04b07 JZ |
210 | * For use when we know that interrupts are disabled, |
211 | * or when we know that preemption is disabled and that | |
212 | * particular counter cannot be updated from interrupt context. | |
2244b95a CL |
213 | */ |
214 | void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item, | |
215 | int delta) | |
216 | { | |
12938a92 CL |
217 | struct per_cpu_pageset __percpu *pcp = zone->pageset; |
218 | s8 __percpu *p = pcp->vm_stat_diff + item; | |
2244b95a | 219 | long x; |
12938a92 CL |
220 | long t; |
221 | ||
222 | x = delta + __this_cpu_read(*p); | |
2244b95a | 223 | |
12938a92 | 224 | t = __this_cpu_read(pcp->stat_threshold); |
2244b95a | 225 | |
12938a92 | 226 | if (unlikely(x > t || x < -t)) { |
2244b95a CL |
227 | zone_page_state_add(x, zone, item); |
228 | x = 0; | |
229 | } | |
12938a92 | 230 | __this_cpu_write(*p, x); |
2244b95a CL |
231 | } |
232 | EXPORT_SYMBOL(__mod_zone_page_state); | |
233 | ||
2244b95a CL |
234 | /* |
235 | * Optimized increment and decrement functions. | |
236 | * | |
237 | * These are only for a single page and therefore can take a struct page * | |
238 | * argument instead of struct zone *. This allows the inclusion of the code | |
239 | * generated for page_zone(page) into the optimized functions. | |
240 | * | |
241 | * No overflow check is necessary and therefore the differential can be | |
242 | * incremented or decremented in place which may allow the compilers to | |
243 | * generate better code. | |
2244b95a CL |
244 | * The increment or decrement is known and therefore one boundary check can |
245 | * be omitted. | |
246 | * | |
df9ecaba CL |
247 | * NOTE: These functions are very performance sensitive. Change only |
248 | * with care. | |
249 | * | |
2244b95a CL |
250 | * Some processors have inc/dec instructions that are atomic vs an interrupt. |
251 | * However, the code must first determine the differential location in a zone | |
252 | * based on the processor number and then inc/dec the counter. There is no | |
253 | * guarantee without disabling preemption that the processor will not change | |
254 | * in between and therefore the atomicity vs. interrupt cannot be exploited | |
255 | * in a useful way here. | |
256 | */ | |
c8785385 | 257 | void __inc_zone_state(struct zone *zone, enum zone_stat_item item) |
2244b95a | 258 | { |
12938a92 CL |
259 | struct per_cpu_pageset __percpu *pcp = zone->pageset; |
260 | s8 __percpu *p = pcp->vm_stat_diff + item; | |
261 | s8 v, t; | |
2244b95a | 262 | |
908ee0f1 | 263 | v = __this_cpu_inc_return(*p); |
12938a92 CL |
264 | t = __this_cpu_read(pcp->stat_threshold); |
265 | if (unlikely(v > t)) { | |
266 | s8 overstep = t >> 1; | |
df9ecaba | 267 | |
12938a92 CL |
268 | zone_page_state_add(v + overstep, zone, item); |
269 | __this_cpu_write(*p, -overstep); | |
2244b95a CL |
270 | } |
271 | } | |
ca889e6c CL |
272 | |
273 | void __inc_zone_page_state(struct page *page, enum zone_stat_item item) | |
274 | { | |
275 | __inc_zone_state(page_zone(page), item); | |
276 | } | |
2244b95a CL |
277 | EXPORT_SYMBOL(__inc_zone_page_state); |
278 | ||
c8785385 | 279 | void __dec_zone_state(struct zone *zone, enum zone_stat_item item) |
2244b95a | 280 | { |
12938a92 CL |
281 | struct per_cpu_pageset __percpu *pcp = zone->pageset; |
282 | s8 __percpu *p = pcp->vm_stat_diff + item; | |
283 | s8 v, t; | |
2244b95a | 284 | |
908ee0f1 | 285 | v = __this_cpu_dec_return(*p); |
12938a92 CL |
286 | t = __this_cpu_read(pcp->stat_threshold); |
287 | if (unlikely(v < - t)) { | |
288 | s8 overstep = t >> 1; | |
2244b95a | 289 | |
12938a92 CL |
290 | zone_page_state_add(v - overstep, zone, item); |
291 | __this_cpu_write(*p, overstep); | |
2244b95a CL |
292 | } |
293 | } | |
c8785385 CL |
294 | |
295 | void __dec_zone_page_state(struct page *page, enum zone_stat_item item) | |
296 | { | |
297 | __dec_zone_state(page_zone(page), item); | |
298 | } | |
2244b95a CL |
299 | EXPORT_SYMBOL(__dec_zone_page_state); |
300 | ||
4156153c | 301 | #ifdef CONFIG_HAVE_CMPXCHG_LOCAL |
7c839120 CL |
302 | /* |
303 | * If we have cmpxchg_local support then we do not need to incur the overhead | |
304 | * that comes with local_irq_save/restore if we use this_cpu_cmpxchg. | |
305 | * | |
306 | * mod_state() modifies the zone counter state through atomic per cpu | |
307 | * operations. | |
308 | * | |
309 | * Overstep mode specifies how overstep should handled: | |
310 | * 0 No overstepping | |
311 | * 1 Overstepping half of threshold | |
312 | * -1 Overstepping minus half of threshold | |
313 | */ | |
314 | static inline void mod_state(struct zone *zone, | |
315 | enum zone_stat_item item, int delta, int overstep_mode) | |
316 | { | |
317 | struct per_cpu_pageset __percpu *pcp = zone->pageset; | |
318 | s8 __percpu *p = pcp->vm_stat_diff + item; | |
319 | long o, n, t, z; | |
320 | ||
321 | do { | |
322 | z = 0; /* overflow to zone counters */ | |
323 | ||
324 | /* | |
325 | * The fetching of the stat_threshold is racy. We may apply | |
326 | * a counter threshold to the wrong the cpu if we get | |
d3bc2367 CL |
327 | * rescheduled while executing here. However, the next |
328 | * counter update will apply the threshold again and | |
329 | * therefore bring the counter under the threshold again. | |
330 | * | |
331 | * Most of the time the thresholds are the same anyways | |
332 | * for all cpus in a zone. | |
7c839120 CL |
333 | */ |
334 | t = this_cpu_read(pcp->stat_threshold); | |
335 | ||
336 | o = this_cpu_read(*p); | |
337 | n = delta + o; | |
338 | ||
339 | if (n > t || n < -t) { | |
340 | int os = overstep_mode * (t >> 1) ; | |
341 | ||
342 | /* Overflow must be added to zone counters */ | |
343 | z = n + os; | |
344 | n = -os; | |
345 | } | |
346 | } while (this_cpu_cmpxchg(*p, o, n) != o); | |
347 | ||
348 | if (z) | |
349 | zone_page_state_add(z, zone, item); | |
350 | } | |
351 | ||
352 | void mod_zone_page_state(struct zone *zone, enum zone_stat_item item, | |
353 | int delta) | |
354 | { | |
355 | mod_state(zone, item, delta, 0); | |
356 | } | |
357 | EXPORT_SYMBOL(mod_zone_page_state); | |
358 | ||
359 | void inc_zone_state(struct zone *zone, enum zone_stat_item item) | |
360 | { | |
361 | mod_state(zone, item, 1, 1); | |
362 | } | |
363 | ||
364 | void inc_zone_page_state(struct page *page, enum zone_stat_item item) | |
365 | { | |
366 | mod_state(page_zone(page), item, 1, 1); | |
367 | } | |
368 | EXPORT_SYMBOL(inc_zone_page_state); | |
369 | ||
370 | void dec_zone_page_state(struct page *page, enum zone_stat_item item) | |
371 | { | |
372 | mod_state(page_zone(page), item, -1, -1); | |
373 | } | |
374 | EXPORT_SYMBOL(dec_zone_page_state); | |
375 | #else | |
376 | /* | |
377 | * Use interrupt disable to serialize counter updates | |
378 | */ | |
379 | void mod_zone_page_state(struct zone *zone, enum zone_stat_item item, | |
380 | int delta) | |
381 | { | |
382 | unsigned long flags; | |
383 | ||
384 | local_irq_save(flags); | |
385 | __mod_zone_page_state(zone, item, delta); | |
386 | local_irq_restore(flags); | |
387 | } | |
388 | EXPORT_SYMBOL(mod_zone_page_state); | |
389 | ||
ca889e6c CL |
390 | void inc_zone_state(struct zone *zone, enum zone_stat_item item) |
391 | { | |
392 | unsigned long flags; | |
393 | ||
394 | local_irq_save(flags); | |
395 | __inc_zone_state(zone, item); | |
396 | local_irq_restore(flags); | |
397 | } | |
398 | ||
2244b95a CL |
399 | void inc_zone_page_state(struct page *page, enum zone_stat_item item) |
400 | { | |
401 | unsigned long flags; | |
402 | struct zone *zone; | |
2244b95a CL |
403 | |
404 | zone = page_zone(page); | |
405 | local_irq_save(flags); | |
ca889e6c | 406 | __inc_zone_state(zone, item); |
2244b95a CL |
407 | local_irq_restore(flags); |
408 | } | |
409 | EXPORT_SYMBOL(inc_zone_page_state); | |
410 | ||
411 | void dec_zone_page_state(struct page *page, enum zone_stat_item item) | |
412 | { | |
413 | unsigned long flags; | |
2244b95a | 414 | |
2244b95a | 415 | local_irq_save(flags); |
a302eb4e | 416 | __dec_zone_page_state(page, item); |
2244b95a CL |
417 | local_irq_restore(flags); |
418 | } | |
419 | EXPORT_SYMBOL(dec_zone_page_state); | |
7c839120 | 420 | #endif |
2244b95a | 421 | |
4edb0748 CL |
422 | static inline void fold_diff(int *diff) |
423 | { | |
424 | int i; | |
425 | ||
426 | for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) | |
427 | if (diff[i]) | |
428 | atomic_long_add(diff[i], &vm_stat[i]); | |
429 | } | |
430 | ||
2244b95a | 431 | /* |
2bb921e5 | 432 | * Update the zone counters for the current cpu. |
a7f75e25 | 433 | * |
4037d452 CL |
434 | * Note that refresh_cpu_vm_stats strives to only access |
435 | * node local memory. The per cpu pagesets on remote zones are placed | |
436 | * in the memory local to the processor using that pageset. So the | |
437 | * loop over all zones will access a series of cachelines local to | |
438 | * the processor. | |
439 | * | |
440 | * The call to zone_page_state_add updates the cachelines with the | |
441 | * statistics in the remote zone struct as well as the global cachelines | |
442 | * with the global counters. These could cause remote node cache line | |
443 | * bouncing and will have to be only done when necessary. | |
2244b95a | 444 | */ |
fbc2edb0 | 445 | static void refresh_cpu_vm_stats(void) |
2244b95a CL |
446 | { |
447 | struct zone *zone; | |
448 | int i; | |
a7f75e25 | 449 | int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, }; |
2244b95a | 450 | |
ee99c71c | 451 | for_each_populated_zone(zone) { |
fbc2edb0 | 452 | struct per_cpu_pageset __percpu *p = zone->pageset; |
2244b95a | 453 | |
fbc2edb0 CL |
454 | for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) { |
455 | int v; | |
2244b95a | 456 | |
fbc2edb0 CL |
457 | v = this_cpu_xchg(p->vm_stat_diff[i], 0); |
458 | if (v) { | |
a7f75e25 | 459 | |
a7f75e25 CL |
460 | atomic_long_add(v, &zone->vm_stat[i]); |
461 | global_diff[i] += v; | |
4037d452 CL |
462 | #ifdef CONFIG_NUMA |
463 | /* 3 seconds idle till flush */ | |
fbc2edb0 | 464 | __this_cpu_write(p->expire, 3); |
4037d452 | 465 | #endif |
2244b95a | 466 | } |
fbc2edb0 | 467 | } |
468fd62e | 468 | cond_resched(); |
4037d452 CL |
469 | #ifdef CONFIG_NUMA |
470 | /* | |
471 | * Deal with draining the remote pageset of this | |
472 | * processor | |
473 | * | |
474 | * Check if there are pages remaining in this pageset | |
475 | * if not then there is nothing to expire. | |
476 | */ | |
fbc2edb0 CL |
477 | if (!__this_cpu_read(p->expire) || |
478 | !__this_cpu_read(p->pcp.count)) | |
4037d452 CL |
479 | continue; |
480 | ||
481 | /* | |
482 | * We never drain zones local to this processor. | |
483 | */ | |
484 | if (zone_to_nid(zone) == numa_node_id()) { | |
fbc2edb0 | 485 | __this_cpu_write(p->expire, 0); |
4037d452 CL |
486 | continue; |
487 | } | |
488 | ||
fbc2edb0 CL |
489 | |
490 | if (__this_cpu_dec_return(p->expire)) | |
4037d452 CL |
491 | continue; |
492 | ||
fbc2edb0 | 493 | if (__this_cpu_read(p->pcp.count)) |
7c8e0181 | 494 | drain_zone_pages(zone, this_cpu_ptr(&p->pcp)); |
4037d452 | 495 | #endif |
2244b95a | 496 | } |
4edb0748 | 497 | fold_diff(global_diff); |
2244b95a CL |
498 | } |
499 | ||
2bb921e5 CL |
500 | /* |
501 | * Fold the data for an offline cpu into the global array. | |
502 | * There cannot be any access by the offline cpu and therefore | |
503 | * synchronization is simplified. | |
504 | */ | |
505 | void cpu_vm_stats_fold(int cpu) | |
506 | { | |
507 | struct zone *zone; | |
508 | int i; | |
509 | int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, }; | |
510 | ||
511 | for_each_populated_zone(zone) { | |
512 | struct per_cpu_pageset *p; | |
513 | ||
514 | p = per_cpu_ptr(zone->pageset, cpu); | |
515 | ||
516 | for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) | |
517 | if (p->vm_stat_diff[i]) { | |
518 | int v; | |
519 | ||
520 | v = p->vm_stat_diff[i]; | |
521 | p->vm_stat_diff[i] = 0; | |
522 | atomic_long_add(v, &zone->vm_stat[i]); | |
523 | global_diff[i] += v; | |
524 | } | |
525 | } | |
526 | ||
4edb0748 | 527 | fold_diff(global_diff); |
2bb921e5 CL |
528 | } |
529 | ||
40f4b1ea CS |
530 | /* |
531 | * this is only called if !populated_zone(zone), which implies no other users of | |
532 | * pset->vm_stat_diff[] exsist. | |
533 | */ | |
5a883813 MK |
534 | void drain_zonestat(struct zone *zone, struct per_cpu_pageset *pset) |
535 | { | |
536 | int i; | |
537 | ||
538 | for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) | |
539 | if (pset->vm_stat_diff[i]) { | |
540 | int v = pset->vm_stat_diff[i]; | |
541 | pset->vm_stat_diff[i] = 0; | |
542 | atomic_long_add(v, &zone->vm_stat[i]); | |
543 | atomic_long_add(v, &vm_stat[i]); | |
544 | } | |
545 | } | |
2244b95a CL |
546 | #endif |
547 | ||
ca889e6c CL |
548 | #ifdef CONFIG_NUMA |
549 | /* | |
550 | * zonelist = the list of zones passed to the allocator | |
551 | * z = the zone from which the allocation occurred. | |
552 | * | |
553 | * Must be called with interrupts disabled. | |
78afd561 AK |
554 | * |
555 | * When __GFP_OTHER_NODE is set assume the node of the preferred | |
556 | * zone is the local node. This is useful for daemons who allocate | |
557 | * memory on behalf of other processes. | |
ca889e6c | 558 | */ |
78afd561 | 559 | void zone_statistics(struct zone *preferred_zone, struct zone *z, gfp_t flags) |
ca889e6c | 560 | { |
18ea7e71 | 561 | if (z->zone_pgdat == preferred_zone->zone_pgdat) { |
ca889e6c CL |
562 | __inc_zone_state(z, NUMA_HIT); |
563 | } else { | |
564 | __inc_zone_state(z, NUMA_MISS); | |
18ea7e71 | 565 | __inc_zone_state(preferred_zone, NUMA_FOREIGN); |
ca889e6c | 566 | } |
78afd561 AK |
567 | if (z->node == ((flags & __GFP_OTHER_NODE) ? |
568 | preferred_zone->node : numa_node_id())) | |
ca889e6c CL |
569 | __inc_zone_state(z, NUMA_LOCAL); |
570 | else | |
571 | __inc_zone_state(z, NUMA_OTHER); | |
572 | } | |
573 | #endif | |
574 | ||
d7a5752c | 575 | #ifdef CONFIG_COMPACTION |
36deb0be | 576 | |
d7a5752c MG |
577 | struct contig_page_info { |
578 | unsigned long free_pages; | |
579 | unsigned long free_blocks_total; | |
580 | unsigned long free_blocks_suitable; | |
581 | }; | |
582 | ||
583 | /* | |
584 | * Calculate the number of free pages in a zone, how many contiguous | |
585 | * pages are free and how many are large enough to satisfy an allocation of | |
586 | * the target size. Note that this function makes no attempt to estimate | |
587 | * how many suitable free blocks there *might* be if MOVABLE pages were | |
588 | * migrated. Calculating that is possible, but expensive and can be | |
589 | * figured out from userspace | |
590 | */ | |
591 | static void fill_contig_page_info(struct zone *zone, | |
592 | unsigned int suitable_order, | |
593 | struct contig_page_info *info) | |
594 | { | |
595 | unsigned int order; | |
596 | ||
597 | info->free_pages = 0; | |
598 | info->free_blocks_total = 0; | |
599 | info->free_blocks_suitable = 0; | |
600 | ||
601 | for (order = 0; order < MAX_ORDER; order++) { | |
602 | unsigned long blocks; | |
603 | ||
604 | /* Count number of free blocks */ | |
605 | blocks = zone->free_area[order].nr_free; | |
606 | info->free_blocks_total += blocks; | |
607 | ||
608 | /* Count free base pages */ | |
609 | info->free_pages += blocks << order; | |
610 | ||
611 | /* Count the suitable free blocks */ | |
612 | if (order >= suitable_order) | |
613 | info->free_blocks_suitable += blocks << | |
614 | (order - suitable_order); | |
615 | } | |
616 | } | |
f1a5ab12 MG |
617 | |
618 | /* | |
619 | * A fragmentation index only makes sense if an allocation of a requested | |
620 | * size would fail. If that is true, the fragmentation index indicates | |
621 | * whether external fragmentation or a lack of memory was the problem. | |
622 | * The value can be used to determine if page reclaim or compaction | |
623 | * should be used | |
624 | */ | |
56de7263 | 625 | static int __fragmentation_index(unsigned int order, struct contig_page_info *info) |
f1a5ab12 MG |
626 | { |
627 | unsigned long requested = 1UL << order; | |
628 | ||
629 | if (!info->free_blocks_total) | |
630 | return 0; | |
631 | ||
632 | /* Fragmentation index only makes sense when a request would fail */ | |
633 | if (info->free_blocks_suitable) | |
634 | return -1000; | |
635 | ||
636 | /* | |
637 | * Index is between 0 and 1 so return within 3 decimal places | |
638 | * | |
639 | * 0 => allocation would fail due to lack of memory | |
640 | * 1 => allocation would fail due to fragmentation | |
641 | */ | |
642 | return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total); | |
643 | } | |
56de7263 MG |
644 | |
645 | /* Same as __fragmentation index but allocs contig_page_info on stack */ | |
646 | int fragmentation_index(struct zone *zone, unsigned int order) | |
647 | { | |
648 | struct contig_page_info info; | |
649 | ||
650 | fill_contig_page_info(zone, order, &info); | |
651 | return __fragmentation_index(order, &info); | |
652 | } | |
d7a5752c MG |
653 | #endif |
654 | ||
655 | #if defined(CONFIG_PROC_FS) || defined(CONFIG_COMPACTION) | |
8f32f7e5 | 656 | #include <linux/proc_fs.h> |
f6ac2354 CL |
657 | #include <linux/seq_file.h> |
658 | ||
467c996c MG |
659 | static char * const migratetype_names[MIGRATE_TYPES] = { |
660 | "Unmovable", | |
661 | "Reclaimable", | |
662 | "Movable", | |
663 | "Reserve", | |
47118af0 MN |
664 | #ifdef CONFIG_CMA |
665 | "CMA", | |
666 | #endif | |
194159fb | 667 | #ifdef CONFIG_MEMORY_ISOLATION |
91446b06 | 668 | "Isolate", |
194159fb | 669 | #endif |
467c996c MG |
670 | }; |
671 | ||
f6ac2354 CL |
672 | static void *frag_start(struct seq_file *m, loff_t *pos) |
673 | { | |
674 | pg_data_t *pgdat; | |
675 | loff_t node = *pos; | |
676 | for (pgdat = first_online_pgdat(); | |
677 | pgdat && node; | |
678 | pgdat = next_online_pgdat(pgdat)) | |
679 | --node; | |
680 | ||
681 | return pgdat; | |
682 | } | |
683 | ||
684 | static void *frag_next(struct seq_file *m, void *arg, loff_t *pos) | |
685 | { | |
686 | pg_data_t *pgdat = (pg_data_t *)arg; | |
687 | ||
688 | (*pos)++; | |
689 | return next_online_pgdat(pgdat); | |
690 | } | |
691 | ||
692 | static void frag_stop(struct seq_file *m, void *arg) | |
693 | { | |
694 | } | |
695 | ||
467c996c MG |
696 | /* Walk all the zones in a node and print using a callback */ |
697 | static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat, | |
698 | void (*print)(struct seq_file *m, pg_data_t *, struct zone *)) | |
f6ac2354 | 699 | { |
f6ac2354 CL |
700 | struct zone *zone; |
701 | struct zone *node_zones = pgdat->node_zones; | |
702 | unsigned long flags; | |
f6ac2354 CL |
703 | |
704 | for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) { | |
705 | if (!populated_zone(zone)) | |
706 | continue; | |
707 | ||
708 | spin_lock_irqsave(&zone->lock, flags); | |
467c996c | 709 | print(m, pgdat, zone); |
f6ac2354 | 710 | spin_unlock_irqrestore(&zone->lock, flags); |
467c996c MG |
711 | } |
712 | } | |
d7a5752c | 713 | #endif |
467c996c | 714 | |
0d6617c7 | 715 | #if defined(CONFIG_PROC_FS) || defined(CONFIG_SYSFS) || defined(CONFIG_NUMA) |
fa25c503 KM |
716 | #ifdef CONFIG_ZONE_DMA |
717 | #define TEXT_FOR_DMA(xx) xx "_dma", | |
718 | #else | |
719 | #define TEXT_FOR_DMA(xx) | |
720 | #endif | |
721 | ||
722 | #ifdef CONFIG_ZONE_DMA32 | |
723 | #define TEXT_FOR_DMA32(xx) xx "_dma32", | |
724 | #else | |
725 | #define TEXT_FOR_DMA32(xx) | |
726 | #endif | |
727 | ||
728 | #ifdef CONFIG_HIGHMEM | |
729 | #define TEXT_FOR_HIGHMEM(xx) xx "_high", | |
730 | #else | |
731 | #define TEXT_FOR_HIGHMEM(xx) | |
732 | #endif | |
733 | ||
734 | #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \ | |
735 | TEXT_FOR_HIGHMEM(xx) xx "_movable", | |
736 | ||
737 | const char * const vmstat_text[] = { | |
738 | /* Zoned VM counters */ | |
739 | "nr_free_pages", | |
81c0a2bb | 740 | "nr_alloc_batch", |
fa25c503 KM |
741 | "nr_inactive_anon", |
742 | "nr_active_anon", | |
743 | "nr_inactive_file", | |
744 | "nr_active_file", | |
745 | "nr_unevictable", | |
746 | "nr_mlock", | |
747 | "nr_anon_pages", | |
748 | "nr_mapped", | |
749 | "nr_file_pages", | |
750 | "nr_dirty", | |
751 | "nr_writeback", | |
752 | "nr_slab_reclaimable", | |
753 | "nr_slab_unreclaimable", | |
754 | "nr_page_table_pages", | |
755 | "nr_kernel_stack", | |
756 | "nr_unstable", | |
757 | "nr_bounce", | |
758 | "nr_vmscan_write", | |
49ea7eb6 | 759 | "nr_vmscan_immediate_reclaim", |
fa25c503 KM |
760 | "nr_writeback_temp", |
761 | "nr_isolated_anon", | |
762 | "nr_isolated_file", | |
763 | "nr_shmem", | |
764 | "nr_dirtied", | |
765 | "nr_written", | |
766 | ||
767 | #ifdef CONFIG_NUMA | |
768 | "numa_hit", | |
769 | "numa_miss", | |
770 | "numa_foreign", | |
771 | "numa_interleave", | |
772 | "numa_local", | |
773 | "numa_other", | |
774 | #endif | |
a528910e JW |
775 | "workingset_refault", |
776 | "workingset_activate", | |
449dd698 | 777 | "workingset_nodereclaim", |
fa25c503 | 778 | "nr_anon_transparent_hugepages", |
d1ce749a | 779 | "nr_free_cma", |
fa25c503 KM |
780 | "nr_dirty_threshold", |
781 | "nr_dirty_background_threshold", | |
782 | ||
783 | #ifdef CONFIG_VM_EVENT_COUNTERS | |
784 | "pgpgin", | |
785 | "pgpgout", | |
786 | "pswpin", | |
787 | "pswpout", | |
788 | ||
789 | TEXTS_FOR_ZONES("pgalloc") | |
790 | ||
791 | "pgfree", | |
792 | "pgactivate", | |
793 | "pgdeactivate", | |
794 | ||
795 | "pgfault", | |
796 | "pgmajfault", | |
797 | ||
798 | TEXTS_FOR_ZONES("pgrefill") | |
904249aa YH |
799 | TEXTS_FOR_ZONES("pgsteal_kswapd") |
800 | TEXTS_FOR_ZONES("pgsteal_direct") | |
fa25c503 KM |
801 | TEXTS_FOR_ZONES("pgscan_kswapd") |
802 | TEXTS_FOR_ZONES("pgscan_direct") | |
68243e76 | 803 | "pgscan_direct_throttle", |
fa25c503 KM |
804 | |
805 | #ifdef CONFIG_NUMA | |
806 | "zone_reclaim_failed", | |
807 | #endif | |
808 | "pginodesteal", | |
809 | "slabs_scanned", | |
fa25c503 KM |
810 | "kswapd_inodesteal", |
811 | "kswapd_low_wmark_hit_quickly", | |
812 | "kswapd_high_wmark_hit_quickly", | |
fa25c503 KM |
813 | "pageoutrun", |
814 | "allocstall", | |
815 | ||
816 | "pgrotated", | |
817 | ||
5509a5d2 DH |
818 | "drop_pagecache", |
819 | "drop_slab", | |
820 | ||
03c5a6e1 MG |
821 | #ifdef CONFIG_NUMA_BALANCING |
822 | "numa_pte_updates", | |
72403b4a | 823 | "numa_huge_pte_updates", |
03c5a6e1 MG |
824 | "numa_hint_faults", |
825 | "numa_hint_faults_local", | |
826 | "numa_pages_migrated", | |
827 | #endif | |
5647bc29 MG |
828 | #ifdef CONFIG_MIGRATION |
829 | "pgmigrate_success", | |
830 | "pgmigrate_fail", | |
831 | #endif | |
fa25c503 | 832 | #ifdef CONFIG_COMPACTION |
397487db MG |
833 | "compact_migrate_scanned", |
834 | "compact_free_scanned", | |
835 | "compact_isolated", | |
fa25c503 KM |
836 | "compact_stall", |
837 | "compact_fail", | |
838 | "compact_success", | |
839 | #endif | |
840 | ||
841 | #ifdef CONFIG_HUGETLB_PAGE | |
842 | "htlb_buddy_alloc_success", | |
843 | "htlb_buddy_alloc_fail", | |
844 | #endif | |
845 | "unevictable_pgs_culled", | |
846 | "unevictable_pgs_scanned", | |
847 | "unevictable_pgs_rescued", | |
848 | "unevictable_pgs_mlocked", | |
849 | "unevictable_pgs_munlocked", | |
850 | "unevictable_pgs_cleared", | |
851 | "unevictable_pgs_stranded", | |
fa25c503 KM |
852 | |
853 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
854 | "thp_fault_alloc", | |
855 | "thp_fault_fallback", | |
856 | "thp_collapse_alloc", | |
857 | "thp_collapse_alloc_failed", | |
858 | "thp_split", | |
d8a8e1f0 KS |
859 | "thp_zero_page_alloc", |
860 | "thp_zero_page_alloc_failed", | |
fa25c503 | 861 | #endif |
ec659934 | 862 | #ifdef CONFIG_DEBUG_TLBFLUSH |
6df46865 | 863 | #ifdef CONFIG_SMP |
9824cf97 DH |
864 | "nr_tlb_remote_flush", |
865 | "nr_tlb_remote_flush_received", | |
ec659934 | 866 | #endif /* CONFIG_SMP */ |
9824cf97 DH |
867 | "nr_tlb_local_flush_all", |
868 | "nr_tlb_local_flush_one", | |
ec659934 | 869 | #endif /* CONFIG_DEBUG_TLBFLUSH */ |
fa25c503 | 870 | |
4f115147 DB |
871 | #ifdef CONFIG_DEBUG_VM_VMACACHE |
872 | "vmacache_find_calls", | |
873 | "vmacache_find_hits", | |
874 | #endif | |
fa25c503 KM |
875 | #endif /* CONFIG_VM_EVENTS_COUNTERS */ |
876 | }; | |
0d6617c7 | 877 | #endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA */ |
fa25c503 KM |
878 | |
879 | ||
d7a5752c | 880 | #ifdef CONFIG_PROC_FS |
467c996c MG |
881 | static void frag_show_print(struct seq_file *m, pg_data_t *pgdat, |
882 | struct zone *zone) | |
883 | { | |
884 | int order; | |
885 | ||
886 | seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name); | |
887 | for (order = 0; order < MAX_ORDER; ++order) | |
888 | seq_printf(m, "%6lu ", zone->free_area[order].nr_free); | |
889 | seq_putc(m, '\n'); | |
890 | } | |
891 | ||
892 | /* | |
893 | * This walks the free areas for each zone. | |
894 | */ | |
895 | static int frag_show(struct seq_file *m, void *arg) | |
896 | { | |
897 | pg_data_t *pgdat = (pg_data_t *)arg; | |
898 | walk_zones_in_node(m, pgdat, frag_show_print); | |
899 | return 0; | |
900 | } | |
901 | ||
902 | static void pagetypeinfo_showfree_print(struct seq_file *m, | |
903 | pg_data_t *pgdat, struct zone *zone) | |
904 | { | |
905 | int order, mtype; | |
906 | ||
907 | for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) { | |
908 | seq_printf(m, "Node %4d, zone %8s, type %12s ", | |
909 | pgdat->node_id, | |
910 | zone->name, | |
911 | migratetype_names[mtype]); | |
912 | for (order = 0; order < MAX_ORDER; ++order) { | |
913 | unsigned long freecount = 0; | |
914 | struct free_area *area; | |
915 | struct list_head *curr; | |
916 | ||
917 | area = &(zone->free_area[order]); | |
918 | ||
919 | list_for_each(curr, &area->free_list[mtype]) | |
920 | freecount++; | |
921 | seq_printf(m, "%6lu ", freecount); | |
922 | } | |
f6ac2354 CL |
923 | seq_putc(m, '\n'); |
924 | } | |
467c996c MG |
925 | } |
926 | ||
927 | /* Print out the free pages at each order for each migatetype */ | |
928 | static int pagetypeinfo_showfree(struct seq_file *m, void *arg) | |
929 | { | |
930 | int order; | |
931 | pg_data_t *pgdat = (pg_data_t *)arg; | |
932 | ||
933 | /* Print header */ | |
934 | seq_printf(m, "%-43s ", "Free pages count per migrate type at order"); | |
935 | for (order = 0; order < MAX_ORDER; ++order) | |
936 | seq_printf(m, "%6d ", order); | |
937 | seq_putc(m, '\n'); | |
938 | ||
939 | walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print); | |
940 | ||
941 | return 0; | |
942 | } | |
943 | ||
944 | static void pagetypeinfo_showblockcount_print(struct seq_file *m, | |
945 | pg_data_t *pgdat, struct zone *zone) | |
946 | { | |
947 | int mtype; | |
948 | unsigned long pfn; | |
949 | unsigned long start_pfn = zone->zone_start_pfn; | |
108bcc96 | 950 | unsigned long end_pfn = zone_end_pfn(zone); |
467c996c MG |
951 | unsigned long count[MIGRATE_TYPES] = { 0, }; |
952 | ||
953 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { | |
954 | struct page *page; | |
955 | ||
956 | if (!pfn_valid(pfn)) | |
957 | continue; | |
958 | ||
959 | page = pfn_to_page(pfn); | |
eb33575c MG |
960 | |
961 | /* Watch for unexpected holes punched in the memmap */ | |
962 | if (!memmap_valid_within(pfn, page, zone)) | |
e80d6a24 | 963 | continue; |
eb33575c | 964 | |
467c996c MG |
965 | mtype = get_pageblock_migratetype(page); |
966 | ||
e80d6a24 MG |
967 | if (mtype < MIGRATE_TYPES) |
968 | count[mtype]++; | |
467c996c MG |
969 | } |
970 | ||
971 | /* Print counts */ | |
972 | seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name); | |
973 | for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) | |
974 | seq_printf(m, "%12lu ", count[mtype]); | |
975 | seq_putc(m, '\n'); | |
976 | } | |
977 | ||
978 | /* Print out the free pages at each order for each migratetype */ | |
979 | static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg) | |
980 | { | |
981 | int mtype; | |
982 | pg_data_t *pgdat = (pg_data_t *)arg; | |
983 | ||
984 | seq_printf(m, "\n%-23s", "Number of blocks type "); | |
985 | for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) | |
986 | seq_printf(m, "%12s ", migratetype_names[mtype]); | |
987 | seq_putc(m, '\n'); | |
988 | walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print); | |
989 | ||
990 | return 0; | |
991 | } | |
992 | ||
993 | /* | |
994 | * This prints out statistics in relation to grouping pages by mobility. | |
995 | * It is expensive to collect so do not constantly read the file. | |
996 | */ | |
997 | static int pagetypeinfo_show(struct seq_file *m, void *arg) | |
998 | { | |
999 | pg_data_t *pgdat = (pg_data_t *)arg; | |
1000 | ||
41b25a37 | 1001 | /* check memoryless node */ |
a47b53c5 | 1002 | if (!node_state(pgdat->node_id, N_MEMORY)) |
41b25a37 KM |
1003 | return 0; |
1004 | ||
467c996c MG |
1005 | seq_printf(m, "Page block order: %d\n", pageblock_order); |
1006 | seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages); | |
1007 | seq_putc(m, '\n'); | |
1008 | pagetypeinfo_showfree(m, pgdat); | |
1009 | pagetypeinfo_showblockcount(m, pgdat); | |
1010 | ||
f6ac2354 CL |
1011 | return 0; |
1012 | } | |
1013 | ||
8f32f7e5 | 1014 | static const struct seq_operations fragmentation_op = { |
f6ac2354 CL |
1015 | .start = frag_start, |
1016 | .next = frag_next, | |
1017 | .stop = frag_stop, | |
1018 | .show = frag_show, | |
1019 | }; | |
1020 | ||
8f32f7e5 AD |
1021 | static int fragmentation_open(struct inode *inode, struct file *file) |
1022 | { | |
1023 | return seq_open(file, &fragmentation_op); | |
1024 | } | |
1025 | ||
1026 | static const struct file_operations fragmentation_file_operations = { | |
1027 | .open = fragmentation_open, | |
1028 | .read = seq_read, | |
1029 | .llseek = seq_lseek, | |
1030 | .release = seq_release, | |
1031 | }; | |
1032 | ||
74e2e8e8 | 1033 | static const struct seq_operations pagetypeinfo_op = { |
467c996c MG |
1034 | .start = frag_start, |
1035 | .next = frag_next, | |
1036 | .stop = frag_stop, | |
1037 | .show = pagetypeinfo_show, | |
1038 | }; | |
1039 | ||
74e2e8e8 AD |
1040 | static int pagetypeinfo_open(struct inode *inode, struct file *file) |
1041 | { | |
1042 | return seq_open(file, &pagetypeinfo_op); | |
1043 | } | |
1044 | ||
1045 | static const struct file_operations pagetypeinfo_file_ops = { | |
1046 | .open = pagetypeinfo_open, | |
1047 | .read = seq_read, | |
1048 | .llseek = seq_lseek, | |
1049 | .release = seq_release, | |
1050 | }; | |
1051 | ||
467c996c MG |
1052 | static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat, |
1053 | struct zone *zone) | |
f6ac2354 | 1054 | { |
467c996c MG |
1055 | int i; |
1056 | seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name); | |
1057 | seq_printf(m, | |
1058 | "\n pages free %lu" | |
1059 | "\n min %lu" | |
1060 | "\n low %lu" | |
1061 | "\n high %lu" | |
08d9ae7c | 1062 | "\n scanned %lu" |
467c996c | 1063 | "\n spanned %lu" |
9feedc9d JL |
1064 | "\n present %lu" |
1065 | "\n managed %lu", | |
88f5acf8 | 1066 | zone_page_state(zone, NR_FREE_PAGES), |
41858966 MG |
1067 | min_wmark_pages(zone), |
1068 | low_wmark_pages(zone), | |
1069 | high_wmark_pages(zone), | |
467c996c | 1070 | zone->pages_scanned, |
467c996c | 1071 | zone->spanned_pages, |
9feedc9d JL |
1072 | zone->present_pages, |
1073 | zone->managed_pages); | |
467c996c MG |
1074 | |
1075 | for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) | |
1076 | seq_printf(m, "\n %-12s %lu", vmstat_text[i], | |
1077 | zone_page_state(zone, i)); | |
1078 | ||
1079 | seq_printf(m, | |
1080 | "\n protection: (%lu", | |
1081 | zone->lowmem_reserve[0]); | |
1082 | for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++) | |
1083 | seq_printf(m, ", %lu", zone->lowmem_reserve[i]); | |
1084 | seq_printf(m, | |
1085 | ")" | |
1086 | "\n pagesets"); | |
1087 | for_each_online_cpu(i) { | |
1088 | struct per_cpu_pageset *pageset; | |
467c996c | 1089 | |
99dcc3e5 | 1090 | pageset = per_cpu_ptr(zone->pageset, i); |
3dfa5721 CL |
1091 | seq_printf(m, |
1092 | "\n cpu: %i" | |
1093 | "\n count: %i" | |
1094 | "\n high: %i" | |
1095 | "\n batch: %i", | |
1096 | i, | |
1097 | pageset->pcp.count, | |
1098 | pageset->pcp.high, | |
1099 | pageset->pcp.batch); | |
df9ecaba | 1100 | #ifdef CONFIG_SMP |
467c996c MG |
1101 | seq_printf(m, "\n vm stats threshold: %d", |
1102 | pageset->stat_threshold); | |
df9ecaba | 1103 | #endif |
f6ac2354 | 1104 | } |
467c996c MG |
1105 | seq_printf(m, |
1106 | "\n all_unreclaimable: %u" | |
556adecb RR |
1107 | "\n start_pfn: %lu" |
1108 | "\n inactive_ratio: %u", | |
6e543d57 | 1109 | !zone_reclaimable(zone), |
556adecb RR |
1110 | zone->zone_start_pfn, |
1111 | zone->inactive_ratio); | |
467c996c MG |
1112 | seq_putc(m, '\n'); |
1113 | } | |
1114 | ||
1115 | /* | |
1116 | * Output information about zones in @pgdat. | |
1117 | */ | |
1118 | static int zoneinfo_show(struct seq_file *m, void *arg) | |
1119 | { | |
1120 | pg_data_t *pgdat = (pg_data_t *)arg; | |
1121 | walk_zones_in_node(m, pgdat, zoneinfo_show_print); | |
f6ac2354 CL |
1122 | return 0; |
1123 | } | |
1124 | ||
5c9fe628 | 1125 | static const struct seq_operations zoneinfo_op = { |
f6ac2354 CL |
1126 | .start = frag_start, /* iterate over all zones. The same as in |
1127 | * fragmentation. */ | |
1128 | .next = frag_next, | |
1129 | .stop = frag_stop, | |
1130 | .show = zoneinfo_show, | |
1131 | }; | |
1132 | ||
5c9fe628 AD |
1133 | static int zoneinfo_open(struct inode *inode, struct file *file) |
1134 | { | |
1135 | return seq_open(file, &zoneinfo_op); | |
1136 | } | |
1137 | ||
1138 | static const struct file_operations proc_zoneinfo_file_operations = { | |
1139 | .open = zoneinfo_open, | |
1140 | .read = seq_read, | |
1141 | .llseek = seq_lseek, | |
1142 | .release = seq_release, | |
1143 | }; | |
1144 | ||
79da826a MR |
1145 | enum writeback_stat_item { |
1146 | NR_DIRTY_THRESHOLD, | |
1147 | NR_DIRTY_BG_THRESHOLD, | |
1148 | NR_VM_WRITEBACK_STAT_ITEMS, | |
1149 | }; | |
1150 | ||
f6ac2354 CL |
1151 | static void *vmstat_start(struct seq_file *m, loff_t *pos) |
1152 | { | |
2244b95a | 1153 | unsigned long *v; |
79da826a | 1154 | int i, stat_items_size; |
f6ac2354 CL |
1155 | |
1156 | if (*pos >= ARRAY_SIZE(vmstat_text)) | |
1157 | return NULL; | |
79da826a MR |
1158 | stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) + |
1159 | NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long); | |
f6ac2354 | 1160 | |
f8891e5e | 1161 | #ifdef CONFIG_VM_EVENT_COUNTERS |
79da826a | 1162 | stat_items_size += sizeof(struct vm_event_state); |
f8891e5e | 1163 | #endif |
79da826a MR |
1164 | |
1165 | v = kmalloc(stat_items_size, GFP_KERNEL); | |
2244b95a CL |
1166 | m->private = v; |
1167 | if (!v) | |
f6ac2354 | 1168 | return ERR_PTR(-ENOMEM); |
2244b95a CL |
1169 | for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) |
1170 | v[i] = global_page_state(i); | |
79da826a MR |
1171 | v += NR_VM_ZONE_STAT_ITEMS; |
1172 | ||
1173 | global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD, | |
1174 | v + NR_DIRTY_THRESHOLD); | |
1175 | v += NR_VM_WRITEBACK_STAT_ITEMS; | |
1176 | ||
f8891e5e | 1177 | #ifdef CONFIG_VM_EVENT_COUNTERS |
79da826a MR |
1178 | all_vm_events(v); |
1179 | v[PGPGIN] /= 2; /* sectors -> kbytes */ | |
1180 | v[PGPGOUT] /= 2; | |
f8891e5e | 1181 | #endif |
ff8b16d7 | 1182 | return (unsigned long *)m->private + *pos; |
f6ac2354 CL |
1183 | } |
1184 | ||
1185 | static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos) | |
1186 | { | |
1187 | (*pos)++; | |
1188 | if (*pos >= ARRAY_SIZE(vmstat_text)) | |
1189 | return NULL; | |
1190 | return (unsigned long *)m->private + *pos; | |
1191 | } | |
1192 | ||
1193 | static int vmstat_show(struct seq_file *m, void *arg) | |
1194 | { | |
1195 | unsigned long *l = arg; | |
1196 | unsigned long off = l - (unsigned long *)m->private; | |
1197 | ||
1198 | seq_printf(m, "%s %lu\n", vmstat_text[off], *l); | |
1199 | return 0; | |
1200 | } | |
1201 | ||
1202 | static void vmstat_stop(struct seq_file *m, void *arg) | |
1203 | { | |
1204 | kfree(m->private); | |
1205 | m->private = NULL; | |
1206 | } | |
1207 | ||
b6aa44ab | 1208 | static const struct seq_operations vmstat_op = { |
f6ac2354 CL |
1209 | .start = vmstat_start, |
1210 | .next = vmstat_next, | |
1211 | .stop = vmstat_stop, | |
1212 | .show = vmstat_show, | |
1213 | }; | |
1214 | ||
b6aa44ab AD |
1215 | static int vmstat_open(struct inode *inode, struct file *file) |
1216 | { | |
1217 | return seq_open(file, &vmstat_op); | |
1218 | } | |
1219 | ||
1220 | static const struct file_operations proc_vmstat_file_operations = { | |
1221 | .open = vmstat_open, | |
1222 | .read = seq_read, | |
1223 | .llseek = seq_lseek, | |
1224 | .release = seq_release, | |
1225 | }; | |
f6ac2354 CL |
1226 | #endif /* CONFIG_PROC_FS */ |
1227 | ||
df9ecaba | 1228 | #ifdef CONFIG_SMP |
d1187ed2 | 1229 | static DEFINE_PER_CPU(struct delayed_work, vmstat_work); |
77461ab3 | 1230 | int sysctl_stat_interval __read_mostly = HZ; |
d1187ed2 CL |
1231 | |
1232 | static void vmstat_update(struct work_struct *w) | |
1233 | { | |
fbc2edb0 | 1234 | refresh_cpu_vm_stats(); |
7c8e0181 | 1235 | schedule_delayed_work(this_cpu_ptr(&vmstat_work), |
98f4ebb2 | 1236 | round_jiffies_relative(sysctl_stat_interval)); |
d1187ed2 CL |
1237 | } |
1238 | ||
0db0628d | 1239 | static void start_cpu_timer(int cpu) |
d1187ed2 | 1240 | { |
1871e52c | 1241 | struct delayed_work *work = &per_cpu(vmstat_work, cpu); |
d1187ed2 | 1242 | |
203b42f7 | 1243 | INIT_DEFERRABLE_WORK(work, vmstat_update); |
1871e52c | 1244 | schedule_delayed_work_on(cpu, work, __round_jiffies_relative(HZ, cpu)); |
d1187ed2 CL |
1245 | } |
1246 | ||
807a1bd2 TK |
1247 | static void vmstat_cpu_dead(int node) |
1248 | { | |
1249 | int cpu; | |
1250 | ||
1251 | get_online_cpus(); | |
1252 | for_each_online_cpu(cpu) | |
1253 | if (cpu_to_node(cpu) == node) | |
1254 | goto end; | |
1255 | ||
1256 | node_clear_state(node, N_CPU); | |
1257 | end: | |
1258 | put_online_cpus(); | |
1259 | } | |
1260 | ||
df9ecaba CL |
1261 | /* |
1262 | * Use the cpu notifier to insure that the thresholds are recalculated | |
1263 | * when necessary. | |
1264 | */ | |
0db0628d | 1265 | static int vmstat_cpuup_callback(struct notifier_block *nfb, |
df9ecaba CL |
1266 | unsigned long action, |
1267 | void *hcpu) | |
1268 | { | |
d1187ed2 CL |
1269 | long cpu = (long)hcpu; |
1270 | ||
df9ecaba | 1271 | switch (action) { |
d1187ed2 CL |
1272 | case CPU_ONLINE: |
1273 | case CPU_ONLINE_FROZEN: | |
5ee28a44 | 1274 | refresh_zone_stat_thresholds(); |
d1187ed2 | 1275 | start_cpu_timer(cpu); |
ad596925 | 1276 | node_set_state(cpu_to_node(cpu), N_CPU); |
d1187ed2 CL |
1277 | break; |
1278 | case CPU_DOWN_PREPARE: | |
1279 | case CPU_DOWN_PREPARE_FROZEN: | |
afe2c511 | 1280 | cancel_delayed_work_sync(&per_cpu(vmstat_work, cpu)); |
d1187ed2 CL |
1281 | per_cpu(vmstat_work, cpu).work.func = NULL; |
1282 | break; | |
1283 | case CPU_DOWN_FAILED: | |
1284 | case CPU_DOWN_FAILED_FROZEN: | |
1285 | start_cpu_timer(cpu); | |
1286 | break; | |
ce421c79 | 1287 | case CPU_DEAD: |
8bb78442 | 1288 | case CPU_DEAD_FROZEN: |
ce421c79 | 1289 | refresh_zone_stat_thresholds(); |
807a1bd2 | 1290 | vmstat_cpu_dead(cpu_to_node(cpu)); |
ce421c79 AW |
1291 | break; |
1292 | default: | |
1293 | break; | |
df9ecaba CL |
1294 | } |
1295 | return NOTIFY_OK; | |
1296 | } | |
1297 | ||
0db0628d | 1298 | static struct notifier_block vmstat_notifier = |
df9ecaba | 1299 | { &vmstat_cpuup_callback, NULL, 0 }; |
8f32f7e5 | 1300 | #endif |
df9ecaba | 1301 | |
e2fc88d0 | 1302 | static int __init setup_vmstat(void) |
df9ecaba | 1303 | { |
8f32f7e5 | 1304 | #ifdef CONFIG_SMP |
d1187ed2 CL |
1305 | int cpu; |
1306 | ||
0be94bad SB |
1307 | cpu_notifier_register_begin(); |
1308 | __register_cpu_notifier(&vmstat_notifier); | |
d1187ed2 | 1309 | |
d7e0b37a | 1310 | for_each_online_cpu(cpu) { |
d1187ed2 | 1311 | start_cpu_timer(cpu); |
d7e0b37a TK |
1312 | node_set_state(cpu_to_node(cpu), N_CPU); |
1313 | } | |
0be94bad | 1314 | cpu_notifier_register_done(); |
8f32f7e5 AD |
1315 | #endif |
1316 | #ifdef CONFIG_PROC_FS | |
1317 | proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations); | |
74e2e8e8 | 1318 | proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops); |
b6aa44ab | 1319 | proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations); |
5c9fe628 | 1320 | proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations); |
8f32f7e5 | 1321 | #endif |
df9ecaba CL |
1322 | return 0; |
1323 | } | |
1324 | module_init(setup_vmstat) | |
d7a5752c MG |
1325 | |
1326 | #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION) | |
1327 | #include <linux/debugfs.h> | |
1328 | ||
d7a5752c MG |
1329 | |
1330 | /* | |
1331 | * Return an index indicating how much of the available free memory is | |
1332 | * unusable for an allocation of the requested size. | |
1333 | */ | |
1334 | static int unusable_free_index(unsigned int order, | |
1335 | struct contig_page_info *info) | |
1336 | { | |
1337 | /* No free memory is interpreted as all free memory is unusable */ | |
1338 | if (info->free_pages == 0) | |
1339 | return 1000; | |
1340 | ||
1341 | /* | |
1342 | * Index should be a value between 0 and 1. Return a value to 3 | |
1343 | * decimal places. | |
1344 | * | |
1345 | * 0 => no fragmentation | |
1346 | * 1 => high fragmentation | |
1347 | */ | |
1348 | return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages); | |
1349 | ||
1350 | } | |
1351 | ||
1352 | static void unusable_show_print(struct seq_file *m, | |
1353 | pg_data_t *pgdat, struct zone *zone) | |
1354 | { | |
1355 | unsigned int order; | |
1356 | int index; | |
1357 | struct contig_page_info info; | |
1358 | ||
1359 | seq_printf(m, "Node %d, zone %8s ", | |
1360 | pgdat->node_id, | |
1361 | zone->name); | |
1362 | for (order = 0; order < MAX_ORDER; ++order) { | |
1363 | fill_contig_page_info(zone, order, &info); | |
1364 | index = unusable_free_index(order, &info); | |
1365 | seq_printf(m, "%d.%03d ", index / 1000, index % 1000); | |
1366 | } | |
1367 | ||
1368 | seq_putc(m, '\n'); | |
1369 | } | |
1370 | ||
1371 | /* | |
1372 | * Display unusable free space index | |
1373 | * | |
1374 | * The unusable free space index measures how much of the available free | |
1375 | * memory cannot be used to satisfy an allocation of a given size and is a | |
1376 | * value between 0 and 1. The higher the value, the more of free memory is | |
1377 | * unusable and by implication, the worse the external fragmentation is. This | |
1378 | * can be expressed as a percentage by multiplying by 100. | |
1379 | */ | |
1380 | static int unusable_show(struct seq_file *m, void *arg) | |
1381 | { | |
1382 | pg_data_t *pgdat = (pg_data_t *)arg; | |
1383 | ||
1384 | /* check memoryless node */ | |
a47b53c5 | 1385 | if (!node_state(pgdat->node_id, N_MEMORY)) |
d7a5752c MG |
1386 | return 0; |
1387 | ||
1388 | walk_zones_in_node(m, pgdat, unusable_show_print); | |
1389 | ||
1390 | return 0; | |
1391 | } | |
1392 | ||
1393 | static const struct seq_operations unusable_op = { | |
1394 | .start = frag_start, | |
1395 | .next = frag_next, | |
1396 | .stop = frag_stop, | |
1397 | .show = unusable_show, | |
1398 | }; | |
1399 | ||
1400 | static int unusable_open(struct inode *inode, struct file *file) | |
1401 | { | |
1402 | return seq_open(file, &unusable_op); | |
1403 | } | |
1404 | ||
1405 | static const struct file_operations unusable_file_ops = { | |
1406 | .open = unusable_open, | |
1407 | .read = seq_read, | |
1408 | .llseek = seq_lseek, | |
1409 | .release = seq_release, | |
1410 | }; | |
1411 | ||
f1a5ab12 MG |
1412 | static void extfrag_show_print(struct seq_file *m, |
1413 | pg_data_t *pgdat, struct zone *zone) | |
1414 | { | |
1415 | unsigned int order; | |
1416 | int index; | |
1417 | ||
1418 | /* Alloc on stack as interrupts are disabled for zone walk */ | |
1419 | struct contig_page_info info; | |
1420 | ||
1421 | seq_printf(m, "Node %d, zone %8s ", | |
1422 | pgdat->node_id, | |
1423 | zone->name); | |
1424 | for (order = 0; order < MAX_ORDER; ++order) { | |
1425 | fill_contig_page_info(zone, order, &info); | |
56de7263 | 1426 | index = __fragmentation_index(order, &info); |
f1a5ab12 MG |
1427 | seq_printf(m, "%d.%03d ", index / 1000, index % 1000); |
1428 | } | |
1429 | ||
1430 | seq_putc(m, '\n'); | |
1431 | } | |
1432 | ||
1433 | /* | |
1434 | * Display fragmentation index for orders that allocations would fail for | |
1435 | */ | |
1436 | static int extfrag_show(struct seq_file *m, void *arg) | |
1437 | { | |
1438 | pg_data_t *pgdat = (pg_data_t *)arg; | |
1439 | ||
1440 | walk_zones_in_node(m, pgdat, extfrag_show_print); | |
1441 | ||
1442 | return 0; | |
1443 | } | |
1444 | ||
1445 | static const struct seq_operations extfrag_op = { | |
1446 | .start = frag_start, | |
1447 | .next = frag_next, | |
1448 | .stop = frag_stop, | |
1449 | .show = extfrag_show, | |
1450 | }; | |
1451 | ||
1452 | static int extfrag_open(struct inode *inode, struct file *file) | |
1453 | { | |
1454 | return seq_open(file, &extfrag_op); | |
1455 | } | |
1456 | ||
1457 | static const struct file_operations extfrag_file_ops = { | |
1458 | .open = extfrag_open, | |
1459 | .read = seq_read, | |
1460 | .llseek = seq_lseek, | |
1461 | .release = seq_release, | |
1462 | }; | |
1463 | ||
d7a5752c MG |
1464 | static int __init extfrag_debug_init(void) |
1465 | { | |
bde8bd8a S |
1466 | struct dentry *extfrag_debug_root; |
1467 | ||
d7a5752c MG |
1468 | extfrag_debug_root = debugfs_create_dir("extfrag", NULL); |
1469 | if (!extfrag_debug_root) | |
1470 | return -ENOMEM; | |
1471 | ||
1472 | if (!debugfs_create_file("unusable_index", 0444, | |
1473 | extfrag_debug_root, NULL, &unusable_file_ops)) | |
bde8bd8a | 1474 | goto fail; |
d7a5752c | 1475 | |
f1a5ab12 MG |
1476 | if (!debugfs_create_file("extfrag_index", 0444, |
1477 | extfrag_debug_root, NULL, &extfrag_file_ops)) | |
bde8bd8a | 1478 | goto fail; |
f1a5ab12 | 1479 | |
d7a5752c | 1480 | return 0; |
bde8bd8a S |
1481 | fail: |
1482 | debugfs_remove_recursive(extfrag_debug_root); | |
1483 | return -ENOMEM; | |
d7a5752c MG |
1484 | } |
1485 | ||
1486 | module_init(extfrag_debug_init); | |
1487 | #endif |