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> |
f6ac2354 | 22 | |
f8891e5e CL |
23 | #ifdef CONFIG_VM_EVENT_COUNTERS |
24 | DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}}; | |
25 | EXPORT_PER_CPU_SYMBOL(vm_event_states); | |
26 | ||
31f961a8 | 27 | static void sum_vm_events(unsigned long *ret) |
f8891e5e | 28 | { |
9eccf2a8 | 29 | int cpu; |
f8891e5e CL |
30 | int i; |
31 | ||
32 | memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long)); | |
33 | ||
31f961a8 | 34 | for_each_online_cpu(cpu) { |
f8891e5e CL |
35 | struct vm_event_state *this = &per_cpu(vm_event_states, cpu); |
36 | ||
f8891e5e CL |
37 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) |
38 | ret[i] += this->event[i]; | |
39 | } | |
40 | } | |
41 | ||
42 | /* | |
43 | * Accumulate the vm event counters across all CPUs. | |
44 | * The result is unavoidably approximate - it can change | |
45 | * during and after execution of this function. | |
46 | */ | |
47 | void all_vm_events(unsigned long *ret) | |
48 | { | |
b5be1132 | 49 | get_online_cpus(); |
31f961a8 | 50 | sum_vm_events(ret); |
b5be1132 | 51 | put_online_cpus(); |
f8891e5e | 52 | } |
32dd66fc | 53 | EXPORT_SYMBOL_GPL(all_vm_events); |
f8891e5e | 54 | |
f8891e5e CL |
55 | /* |
56 | * Fold the foreign cpu events into our own. | |
57 | * | |
58 | * This is adding to the events on one processor | |
59 | * but keeps the global counts constant. | |
60 | */ | |
61 | void vm_events_fold_cpu(int cpu) | |
62 | { | |
63 | struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu); | |
64 | int i; | |
65 | ||
66 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { | |
67 | count_vm_events(i, fold_state->event[i]); | |
68 | fold_state->event[i] = 0; | |
69 | } | |
70 | } | |
f8891e5e CL |
71 | |
72 | #endif /* CONFIG_VM_EVENT_COUNTERS */ | |
73 | ||
2244b95a CL |
74 | /* |
75 | * Manage combined zone based / global counters | |
76 | * | |
77 | * vm_stat contains the global counters | |
78 | */ | |
a1cb2c60 | 79 | atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS] __cacheline_aligned_in_smp; |
2244b95a CL |
80 | EXPORT_SYMBOL(vm_stat); |
81 | ||
82 | #ifdef CONFIG_SMP | |
83 | ||
b44129b3 | 84 | int calculate_pressure_threshold(struct zone *zone) |
88f5acf8 MG |
85 | { |
86 | int threshold; | |
87 | int watermark_distance; | |
88 | ||
89 | /* | |
90 | * As vmstats are not up to date, there is drift between the estimated | |
91 | * and real values. For high thresholds and a high number of CPUs, it | |
92 | * is possible for the min watermark to be breached while the estimated | |
93 | * value looks fine. The pressure threshold is a reduced value such | |
94 | * that even the maximum amount of drift will not accidentally breach | |
95 | * the min watermark | |
96 | */ | |
97 | watermark_distance = low_wmark_pages(zone) - min_wmark_pages(zone); | |
98 | threshold = max(1, (int)(watermark_distance / num_online_cpus())); | |
99 | ||
100 | /* | |
101 | * Maximum threshold is 125 | |
102 | */ | |
103 | threshold = min(125, threshold); | |
104 | ||
105 | return threshold; | |
106 | } | |
107 | ||
b44129b3 | 108 | int calculate_normal_threshold(struct zone *zone) |
df9ecaba CL |
109 | { |
110 | int threshold; | |
111 | int mem; /* memory in 128 MB units */ | |
112 | ||
113 | /* | |
114 | * The threshold scales with the number of processors and the amount | |
115 | * of memory per zone. More memory means that we can defer updates for | |
116 | * longer, more processors could lead to more contention. | |
117 | * fls() is used to have a cheap way of logarithmic scaling. | |
118 | * | |
119 | * Some sample thresholds: | |
120 | * | |
121 | * Threshold Processors (fls) Zonesize fls(mem+1) | |
122 | * ------------------------------------------------------------------ | |
123 | * 8 1 1 0.9-1 GB 4 | |
124 | * 16 2 2 0.9-1 GB 4 | |
125 | * 20 2 2 1-2 GB 5 | |
126 | * 24 2 2 2-4 GB 6 | |
127 | * 28 2 2 4-8 GB 7 | |
128 | * 32 2 2 8-16 GB 8 | |
129 | * 4 2 2 <128M 1 | |
130 | * 30 4 3 2-4 GB 5 | |
131 | * 48 4 3 8-16 GB 8 | |
132 | * 32 8 4 1-2 GB 4 | |
133 | * 32 8 4 0.9-1GB 4 | |
134 | * 10 16 5 <128M 1 | |
135 | * 40 16 5 900M 4 | |
136 | * 70 64 7 2-4 GB 5 | |
137 | * 84 64 7 4-8 GB 6 | |
138 | * 108 512 9 4-8 GB 6 | |
139 | * 125 1024 10 8-16 GB 8 | |
140 | * 125 1024 10 16-32 GB 9 | |
141 | */ | |
142 | ||
b40da049 | 143 | mem = zone->managed_pages >> (27 - PAGE_SHIFT); |
df9ecaba CL |
144 | |
145 | threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem)); | |
146 | ||
147 | /* | |
148 | * Maximum threshold is 125 | |
149 | */ | |
150 | threshold = min(125, threshold); | |
151 | ||
152 | return threshold; | |
153 | } | |
2244b95a CL |
154 | |
155 | /* | |
df9ecaba | 156 | * Refresh the thresholds for each zone. |
2244b95a | 157 | */ |
a6cccdc3 | 158 | void refresh_zone_stat_thresholds(void) |
2244b95a | 159 | { |
df9ecaba CL |
160 | struct zone *zone; |
161 | int cpu; | |
162 | int threshold; | |
163 | ||
ee99c71c | 164 | for_each_populated_zone(zone) { |
aa454840 CL |
165 | unsigned long max_drift, tolerate_drift; |
166 | ||
b44129b3 | 167 | threshold = calculate_normal_threshold(zone); |
df9ecaba CL |
168 | |
169 | for_each_online_cpu(cpu) | |
99dcc3e5 CL |
170 | per_cpu_ptr(zone->pageset, cpu)->stat_threshold |
171 | = threshold; | |
aa454840 CL |
172 | |
173 | /* | |
174 | * Only set percpu_drift_mark if there is a danger that | |
175 | * NR_FREE_PAGES reports the low watermark is ok when in fact | |
176 | * the min watermark could be breached by an allocation | |
177 | */ | |
178 | tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone); | |
179 | max_drift = num_online_cpus() * threshold; | |
180 | if (max_drift > tolerate_drift) | |
181 | zone->percpu_drift_mark = high_wmark_pages(zone) + | |
182 | max_drift; | |
df9ecaba | 183 | } |
2244b95a CL |
184 | } |
185 | ||
b44129b3 MG |
186 | void set_pgdat_percpu_threshold(pg_data_t *pgdat, |
187 | int (*calculate_pressure)(struct zone *)) | |
88f5acf8 MG |
188 | { |
189 | struct zone *zone; | |
190 | int cpu; | |
191 | int threshold; | |
192 | int i; | |
193 | ||
88f5acf8 MG |
194 | for (i = 0; i < pgdat->nr_zones; i++) { |
195 | zone = &pgdat->node_zones[i]; | |
196 | if (!zone->percpu_drift_mark) | |
197 | continue; | |
198 | ||
b44129b3 MG |
199 | threshold = (*calculate_pressure)(zone); |
200 | for_each_possible_cpu(cpu) | |
88f5acf8 MG |
201 | per_cpu_ptr(zone->pageset, cpu)->stat_threshold |
202 | = threshold; | |
203 | } | |
88f5acf8 MG |
204 | } |
205 | ||
2244b95a CL |
206 | /* |
207 | * For use when we know that interrupts are disabled. | |
208 | */ | |
209 | void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item, | |
210 | int delta) | |
211 | { | |
12938a92 CL |
212 | struct per_cpu_pageset __percpu *pcp = zone->pageset; |
213 | s8 __percpu *p = pcp->vm_stat_diff + item; | |
2244b95a | 214 | long x; |
12938a92 CL |
215 | long t; |
216 | ||
217 | x = delta + __this_cpu_read(*p); | |
2244b95a | 218 | |
12938a92 | 219 | t = __this_cpu_read(pcp->stat_threshold); |
2244b95a | 220 | |
12938a92 | 221 | if (unlikely(x > t || x < -t)) { |
2244b95a CL |
222 | zone_page_state_add(x, zone, item); |
223 | x = 0; | |
224 | } | |
12938a92 | 225 | __this_cpu_write(*p, x); |
2244b95a CL |
226 | } |
227 | EXPORT_SYMBOL(__mod_zone_page_state); | |
228 | ||
2244b95a CL |
229 | /* |
230 | * Optimized increment and decrement functions. | |
231 | * | |
232 | * These are only for a single page and therefore can take a struct page * | |
233 | * argument instead of struct zone *. This allows the inclusion of the code | |
234 | * generated for page_zone(page) into the optimized functions. | |
235 | * | |
236 | * No overflow check is necessary and therefore the differential can be | |
237 | * incremented or decremented in place which may allow the compilers to | |
238 | * generate better code. | |
2244b95a CL |
239 | * The increment or decrement is known and therefore one boundary check can |
240 | * be omitted. | |
241 | * | |
df9ecaba CL |
242 | * NOTE: These functions are very performance sensitive. Change only |
243 | * with care. | |
244 | * | |
2244b95a CL |
245 | * Some processors have inc/dec instructions that are atomic vs an interrupt. |
246 | * However, the code must first determine the differential location in a zone | |
247 | * based on the processor number and then inc/dec the counter. There is no | |
248 | * guarantee without disabling preemption that the processor will not change | |
249 | * in between and therefore the atomicity vs. interrupt cannot be exploited | |
250 | * in a useful way here. | |
251 | */ | |
c8785385 | 252 | void __inc_zone_state(struct zone *zone, enum zone_stat_item item) |
2244b95a | 253 | { |
12938a92 CL |
254 | struct per_cpu_pageset __percpu *pcp = zone->pageset; |
255 | s8 __percpu *p = pcp->vm_stat_diff + item; | |
256 | s8 v, t; | |
2244b95a | 257 | |
908ee0f1 | 258 | v = __this_cpu_inc_return(*p); |
12938a92 CL |
259 | t = __this_cpu_read(pcp->stat_threshold); |
260 | if (unlikely(v > t)) { | |
261 | s8 overstep = t >> 1; | |
df9ecaba | 262 | |
12938a92 CL |
263 | zone_page_state_add(v + overstep, zone, item); |
264 | __this_cpu_write(*p, -overstep); | |
2244b95a CL |
265 | } |
266 | } | |
ca889e6c CL |
267 | |
268 | void __inc_zone_page_state(struct page *page, enum zone_stat_item item) | |
269 | { | |
270 | __inc_zone_state(page_zone(page), item); | |
271 | } | |
2244b95a CL |
272 | EXPORT_SYMBOL(__inc_zone_page_state); |
273 | ||
c8785385 | 274 | void __dec_zone_state(struct zone *zone, enum zone_stat_item item) |
2244b95a | 275 | { |
12938a92 CL |
276 | struct per_cpu_pageset __percpu *pcp = zone->pageset; |
277 | s8 __percpu *p = pcp->vm_stat_diff + item; | |
278 | s8 v, t; | |
2244b95a | 279 | |
908ee0f1 | 280 | v = __this_cpu_dec_return(*p); |
12938a92 CL |
281 | t = __this_cpu_read(pcp->stat_threshold); |
282 | if (unlikely(v < - t)) { | |
283 | s8 overstep = t >> 1; | |
2244b95a | 284 | |
12938a92 CL |
285 | zone_page_state_add(v - overstep, zone, item); |
286 | __this_cpu_write(*p, overstep); | |
2244b95a CL |
287 | } |
288 | } | |
c8785385 CL |
289 | |
290 | void __dec_zone_page_state(struct page *page, enum zone_stat_item item) | |
291 | { | |
292 | __dec_zone_state(page_zone(page), item); | |
293 | } | |
2244b95a CL |
294 | EXPORT_SYMBOL(__dec_zone_page_state); |
295 | ||
4156153c | 296 | #ifdef CONFIG_HAVE_CMPXCHG_LOCAL |
7c839120 CL |
297 | /* |
298 | * If we have cmpxchg_local support then we do not need to incur the overhead | |
299 | * that comes with local_irq_save/restore if we use this_cpu_cmpxchg. | |
300 | * | |
301 | * mod_state() modifies the zone counter state through atomic per cpu | |
302 | * operations. | |
303 | * | |
304 | * Overstep mode specifies how overstep should handled: | |
305 | * 0 No overstepping | |
306 | * 1 Overstepping half of threshold | |
307 | * -1 Overstepping minus half of threshold | |
308 | */ | |
309 | static inline void mod_state(struct zone *zone, | |
310 | enum zone_stat_item item, int delta, int overstep_mode) | |
311 | { | |
312 | struct per_cpu_pageset __percpu *pcp = zone->pageset; | |
313 | s8 __percpu *p = pcp->vm_stat_diff + item; | |
314 | long o, n, t, z; | |
315 | ||
316 | do { | |
317 | z = 0; /* overflow to zone counters */ | |
318 | ||
319 | /* | |
320 | * The fetching of the stat_threshold is racy. We may apply | |
321 | * a counter threshold to the wrong the cpu if we get | |
d3bc2367 CL |
322 | * rescheduled while executing here. However, the next |
323 | * counter update will apply the threshold again and | |
324 | * therefore bring the counter under the threshold again. | |
325 | * | |
326 | * Most of the time the thresholds are the same anyways | |
327 | * for all cpus in a zone. | |
7c839120 CL |
328 | */ |
329 | t = this_cpu_read(pcp->stat_threshold); | |
330 | ||
331 | o = this_cpu_read(*p); | |
332 | n = delta + o; | |
333 | ||
334 | if (n > t || n < -t) { | |
335 | int os = overstep_mode * (t >> 1) ; | |
336 | ||
337 | /* Overflow must be added to zone counters */ | |
338 | z = n + os; | |
339 | n = -os; | |
340 | } | |
341 | } while (this_cpu_cmpxchg(*p, o, n) != o); | |
342 | ||
343 | if (z) | |
344 | zone_page_state_add(z, zone, item); | |
345 | } | |
346 | ||
347 | void mod_zone_page_state(struct zone *zone, enum zone_stat_item item, | |
348 | int delta) | |
349 | { | |
350 | mod_state(zone, item, delta, 0); | |
351 | } | |
352 | EXPORT_SYMBOL(mod_zone_page_state); | |
353 | ||
354 | void inc_zone_state(struct zone *zone, enum zone_stat_item item) | |
355 | { | |
356 | mod_state(zone, item, 1, 1); | |
357 | } | |
358 | ||
359 | void inc_zone_page_state(struct page *page, enum zone_stat_item item) | |
360 | { | |
361 | mod_state(page_zone(page), item, 1, 1); | |
362 | } | |
363 | EXPORT_SYMBOL(inc_zone_page_state); | |
364 | ||
365 | void dec_zone_page_state(struct page *page, enum zone_stat_item item) | |
366 | { | |
367 | mod_state(page_zone(page), item, -1, -1); | |
368 | } | |
369 | EXPORT_SYMBOL(dec_zone_page_state); | |
370 | #else | |
371 | /* | |
372 | * Use interrupt disable to serialize counter updates | |
373 | */ | |
374 | void mod_zone_page_state(struct zone *zone, enum zone_stat_item item, | |
375 | int delta) | |
376 | { | |
377 | unsigned long flags; | |
378 | ||
379 | local_irq_save(flags); | |
380 | __mod_zone_page_state(zone, item, delta); | |
381 | local_irq_restore(flags); | |
382 | } | |
383 | EXPORT_SYMBOL(mod_zone_page_state); | |
384 | ||
ca889e6c CL |
385 | void inc_zone_state(struct zone *zone, enum zone_stat_item item) |
386 | { | |
387 | unsigned long flags; | |
388 | ||
389 | local_irq_save(flags); | |
390 | __inc_zone_state(zone, item); | |
391 | local_irq_restore(flags); | |
392 | } | |
393 | ||
2244b95a CL |
394 | void inc_zone_page_state(struct page *page, enum zone_stat_item item) |
395 | { | |
396 | unsigned long flags; | |
397 | struct zone *zone; | |
2244b95a CL |
398 | |
399 | zone = page_zone(page); | |
400 | local_irq_save(flags); | |
ca889e6c | 401 | __inc_zone_state(zone, item); |
2244b95a CL |
402 | local_irq_restore(flags); |
403 | } | |
404 | EXPORT_SYMBOL(inc_zone_page_state); | |
405 | ||
406 | void dec_zone_page_state(struct page *page, enum zone_stat_item item) | |
407 | { | |
408 | unsigned long flags; | |
2244b95a | 409 | |
2244b95a | 410 | local_irq_save(flags); |
a302eb4e | 411 | __dec_zone_page_state(page, item); |
2244b95a CL |
412 | local_irq_restore(flags); |
413 | } | |
414 | EXPORT_SYMBOL(dec_zone_page_state); | |
7c839120 | 415 | #endif |
2244b95a CL |
416 | |
417 | /* | |
418 | * Update the zone counters for one cpu. | |
4037d452 | 419 | * |
a7f75e25 CL |
420 | * The cpu specified must be either the current cpu or a processor that |
421 | * is not online. If it is the current cpu then the execution thread must | |
422 | * be pinned to the current cpu. | |
423 | * | |
4037d452 CL |
424 | * Note that refresh_cpu_vm_stats strives to only access |
425 | * node local memory. The per cpu pagesets on remote zones are placed | |
426 | * in the memory local to the processor using that pageset. So the | |
427 | * loop over all zones will access a series of cachelines local to | |
428 | * the processor. | |
429 | * | |
430 | * The call to zone_page_state_add updates the cachelines with the | |
431 | * statistics in the remote zone struct as well as the global cachelines | |
432 | * with the global counters. These could cause remote node cache line | |
433 | * bouncing and will have to be only done when necessary. | |
2244b95a CL |
434 | */ |
435 | void refresh_cpu_vm_stats(int cpu) | |
436 | { | |
437 | struct zone *zone; | |
438 | int i; | |
a7f75e25 | 439 | int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, }; |
2244b95a | 440 | |
ee99c71c | 441 | for_each_populated_zone(zone) { |
4037d452 | 442 | struct per_cpu_pageset *p; |
2244b95a | 443 | |
99dcc3e5 | 444 | p = per_cpu_ptr(zone->pageset, cpu); |
2244b95a CL |
445 | |
446 | for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) | |
4037d452 | 447 | if (p->vm_stat_diff[i]) { |
a7f75e25 CL |
448 | unsigned long flags; |
449 | int v; | |
450 | ||
2244b95a | 451 | local_irq_save(flags); |
a7f75e25 | 452 | v = p->vm_stat_diff[i]; |
4037d452 | 453 | p->vm_stat_diff[i] = 0; |
a7f75e25 CL |
454 | local_irq_restore(flags); |
455 | atomic_long_add(v, &zone->vm_stat[i]); | |
456 | global_diff[i] += v; | |
4037d452 CL |
457 | #ifdef CONFIG_NUMA |
458 | /* 3 seconds idle till flush */ | |
459 | p->expire = 3; | |
460 | #endif | |
2244b95a | 461 | } |
468fd62e | 462 | cond_resched(); |
4037d452 CL |
463 | #ifdef CONFIG_NUMA |
464 | /* | |
465 | * Deal with draining the remote pageset of this | |
466 | * processor | |
467 | * | |
468 | * Check if there are pages remaining in this pageset | |
469 | * if not then there is nothing to expire. | |
470 | */ | |
3dfa5721 | 471 | if (!p->expire || !p->pcp.count) |
4037d452 CL |
472 | continue; |
473 | ||
474 | /* | |
475 | * We never drain zones local to this processor. | |
476 | */ | |
477 | if (zone_to_nid(zone) == numa_node_id()) { | |
478 | p->expire = 0; | |
479 | continue; | |
480 | } | |
481 | ||
482 | p->expire--; | |
483 | if (p->expire) | |
484 | continue; | |
485 | ||
3dfa5721 CL |
486 | if (p->pcp.count) |
487 | drain_zone_pages(zone, &p->pcp); | |
4037d452 | 488 | #endif |
2244b95a | 489 | } |
a7f75e25 CL |
490 | |
491 | for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) | |
492 | if (global_diff[i]) | |
493 | atomic_long_add(global_diff[i], &vm_stat[i]); | |
2244b95a CL |
494 | } |
495 | ||
5a883813 MK |
496 | void drain_zonestat(struct zone *zone, struct per_cpu_pageset *pset) |
497 | { | |
498 | int i; | |
499 | ||
500 | for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) | |
501 | if (pset->vm_stat_diff[i]) { | |
502 | int v = pset->vm_stat_diff[i]; | |
503 | pset->vm_stat_diff[i] = 0; | |
504 | atomic_long_add(v, &zone->vm_stat[i]); | |
505 | atomic_long_add(v, &vm_stat[i]); | |
506 | } | |
507 | } | |
2244b95a CL |
508 | #endif |
509 | ||
ca889e6c CL |
510 | #ifdef CONFIG_NUMA |
511 | /* | |
512 | * zonelist = the list of zones passed to the allocator | |
513 | * z = the zone from which the allocation occurred. | |
514 | * | |
515 | * Must be called with interrupts disabled. | |
78afd561 AK |
516 | * |
517 | * When __GFP_OTHER_NODE is set assume the node of the preferred | |
518 | * zone is the local node. This is useful for daemons who allocate | |
519 | * memory on behalf of other processes. | |
ca889e6c | 520 | */ |
78afd561 | 521 | void zone_statistics(struct zone *preferred_zone, struct zone *z, gfp_t flags) |
ca889e6c | 522 | { |
18ea7e71 | 523 | if (z->zone_pgdat == preferred_zone->zone_pgdat) { |
ca889e6c CL |
524 | __inc_zone_state(z, NUMA_HIT); |
525 | } else { | |
526 | __inc_zone_state(z, NUMA_MISS); | |
18ea7e71 | 527 | __inc_zone_state(preferred_zone, NUMA_FOREIGN); |
ca889e6c | 528 | } |
78afd561 AK |
529 | if (z->node == ((flags & __GFP_OTHER_NODE) ? |
530 | preferred_zone->node : numa_node_id())) | |
ca889e6c CL |
531 | __inc_zone_state(z, NUMA_LOCAL); |
532 | else | |
533 | __inc_zone_state(z, NUMA_OTHER); | |
534 | } | |
535 | #endif | |
536 | ||
d7a5752c | 537 | #ifdef CONFIG_COMPACTION |
36deb0be | 538 | |
d7a5752c MG |
539 | struct contig_page_info { |
540 | unsigned long free_pages; | |
541 | unsigned long free_blocks_total; | |
542 | unsigned long free_blocks_suitable; | |
543 | }; | |
544 | ||
545 | /* | |
546 | * Calculate the number of free pages in a zone, how many contiguous | |
547 | * pages are free and how many are large enough to satisfy an allocation of | |
548 | * the target size. Note that this function makes no attempt to estimate | |
549 | * how many suitable free blocks there *might* be if MOVABLE pages were | |
550 | * migrated. Calculating that is possible, but expensive and can be | |
551 | * figured out from userspace | |
552 | */ | |
553 | static void fill_contig_page_info(struct zone *zone, | |
554 | unsigned int suitable_order, | |
555 | struct contig_page_info *info) | |
556 | { | |
557 | unsigned int order; | |
558 | ||
559 | info->free_pages = 0; | |
560 | info->free_blocks_total = 0; | |
561 | info->free_blocks_suitable = 0; | |
562 | ||
563 | for (order = 0; order < MAX_ORDER; order++) { | |
564 | unsigned long blocks; | |
565 | ||
566 | /* Count number of free blocks */ | |
567 | blocks = zone->free_area[order].nr_free; | |
568 | info->free_blocks_total += blocks; | |
569 | ||
570 | /* Count free base pages */ | |
571 | info->free_pages += blocks << order; | |
572 | ||
573 | /* Count the suitable free blocks */ | |
574 | if (order >= suitable_order) | |
575 | info->free_blocks_suitable += blocks << | |
576 | (order - suitable_order); | |
577 | } | |
578 | } | |
f1a5ab12 MG |
579 | |
580 | /* | |
581 | * A fragmentation index only makes sense if an allocation of a requested | |
582 | * size would fail. If that is true, the fragmentation index indicates | |
583 | * whether external fragmentation or a lack of memory was the problem. | |
584 | * The value can be used to determine if page reclaim or compaction | |
585 | * should be used | |
586 | */ | |
56de7263 | 587 | static int __fragmentation_index(unsigned int order, struct contig_page_info *info) |
f1a5ab12 MG |
588 | { |
589 | unsigned long requested = 1UL << order; | |
590 | ||
591 | if (!info->free_blocks_total) | |
592 | return 0; | |
593 | ||
594 | /* Fragmentation index only makes sense when a request would fail */ | |
595 | if (info->free_blocks_suitable) | |
596 | return -1000; | |
597 | ||
598 | /* | |
599 | * Index is between 0 and 1 so return within 3 decimal places | |
600 | * | |
601 | * 0 => allocation would fail due to lack of memory | |
602 | * 1 => allocation would fail due to fragmentation | |
603 | */ | |
604 | return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total); | |
605 | } | |
56de7263 MG |
606 | |
607 | /* Same as __fragmentation index but allocs contig_page_info on stack */ | |
608 | int fragmentation_index(struct zone *zone, unsigned int order) | |
609 | { | |
610 | struct contig_page_info info; | |
611 | ||
612 | fill_contig_page_info(zone, order, &info); | |
613 | return __fragmentation_index(order, &info); | |
614 | } | |
d7a5752c MG |
615 | #endif |
616 | ||
617 | #if defined(CONFIG_PROC_FS) || defined(CONFIG_COMPACTION) | |
8f32f7e5 | 618 | #include <linux/proc_fs.h> |
f6ac2354 CL |
619 | #include <linux/seq_file.h> |
620 | ||
467c996c MG |
621 | static char * const migratetype_names[MIGRATE_TYPES] = { |
622 | "Unmovable", | |
623 | "Reclaimable", | |
624 | "Movable", | |
625 | "Reserve", | |
47118af0 MN |
626 | #ifdef CONFIG_CMA |
627 | "CMA", | |
628 | #endif | |
194159fb | 629 | #ifdef CONFIG_MEMORY_ISOLATION |
91446b06 | 630 | "Isolate", |
194159fb | 631 | #endif |
467c996c MG |
632 | }; |
633 | ||
f6ac2354 CL |
634 | static void *frag_start(struct seq_file *m, loff_t *pos) |
635 | { | |
636 | pg_data_t *pgdat; | |
637 | loff_t node = *pos; | |
638 | for (pgdat = first_online_pgdat(); | |
639 | pgdat && node; | |
640 | pgdat = next_online_pgdat(pgdat)) | |
641 | --node; | |
642 | ||
643 | return pgdat; | |
644 | } | |
645 | ||
646 | static void *frag_next(struct seq_file *m, void *arg, loff_t *pos) | |
647 | { | |
648 | pg_data_t *pgdat = (pg_data_t *)arg; | |
649 | ||
650 | (*pos)++; | |
651 | return next_online_pgdat(pgdat); | |
652 | } | |
653 | ||
654 | static void frag_stop(struct seq_file *m, void *arg) | |
655 | { | |
656 | } | |
657 | ||
467c996c MG |
658 | /* Walk all the zones in a node and print using a callback */ |
659 | static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat, | |
660 | void (*print)(struct seq_file *m, pg_data_t *, struct zone *)) | |
f6ac2354 | 661 | { |
f6ac2354 CL |
662 | struct zone *zone; |
663 | struct zone *node_zones = pgdat->node_zones; | |
664 | unsigned long flags; | |
f6ac2354 CL |
665 | |
666 | for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) { | |
667 | if (!populated_zone(zone)) | |
668 | continue; | |
669 | ||
670 | spin_lock_irqsave(&zone->lock, flags); | |
467c996c | 671 | print(m, pgdat, zone); |
f6ac2354 | 672 | spin_unlock_irqrestore(&zone->lock, flags); |
467c996c MG |
673 | } |
674 | } | |
d7a5752c | 675 | #endif |
467c996c | 676 | |
0d6617c7 | 677 | #if defined(CONFIG_PROC_FS) || defined(CONFIG_SYSFS) || defined(CONFIG_NUMA) |
fa25c503 KM |
678 | #ifdef CONFIG_ZONE_DMA |
679 | #define TEXT_FOR_DMA(xx) xx "_dma", | |
680 | #else | |
681 | #define TEXT_FOR_DMA(xx) | |
682 | #endif | |
683 | ||
684 | #ifdef CONFIG_ZONE_DMA32 | |
685 | #define TEXT_FOR_DMA32(xx) xx "_dma32", | |
686 | #else | |
687 | #define TEXT_FOR_DMA32(xx) | |
688 | #endif | |
689 | ||
690 | #ifdef CONFIG_HIGHMEM | |
691 | #define TEXT_FOR_HIGHMEM(xx) xx "_high", | |
692 | #else | |
693 | #define TEXT_FOR_HIGHMEM(xx) | |
694 | #endif | |
695 | ||
696 | #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \ | |
697 | TEXT_FOR_HIGHMEM(xx) xx "_movable", | |
698 | ||
699 | const char * const vmstat_text[] = { | |
700 | /* Zoned VM counters */ | |
701 | "nr_free_pages", | |
702 | "nr_inactive_anon", | |
703 | "nr_active_anon", | |
704 | "nr_inactive_file", | |
705 | "nr_active_file", | |
706 | "nr_unevictable", | |
707 | "nr_mlock", | |
708 | "nr_anon_pages", | |
709 | "nr_mapped", | |
710 | "nr_file_pages", | |
711 | "nr_dirty", | |
712 | "nr_writeback", | |
713 | "nr_slab_reclaimable", | |
714 | "nr_slab_unreclaimable", | |
715 | "nr_page_table_pages", | |
716 | "nr_kernel_stack", | |
717 | "nr_unstable", | |
718 | "nr_bounce", | |
719 | "nr_vmscan_write", | |
49ea7eb6 | 720 | "nr_vmscan_immediate_reclaim", |
fa25c503 KM |
721 | "nr_writeback_temp", |
722 | "nr_isolated_anon", | |
723 | "nr_isolated_file", | |
724 | "nr_shmem", | |
725 | "nr_dirtied", | |
726 | "nr_written", | |
727 | ||
728 | #ifdef CONFIG_NUMA | |
729 | "numa_hit", | |
730 | "numa_miss", | |
731 | "numa_foreign", | |
732 | "numa_interleave", | |
733 | "numa_local", | |
734 | "numa_other", | |
735 | #endif | |
736 | "nr_anon_transparent_hugepages", | |
d1ce749a | 737 | "nr_free_cma", |
fa25c503 KM |
738 | "nr_dirty_threshold", |
739 | "nr_dirty_background_threshold", | |
740 | ||
741 | #ifdef CONFIG_VM_EVENT_COUNTERS | |
742 | "pgpgin", | |
743 | "pgpgout", | |
744 | "pswpin", | |
745 | "pswpout", | |
746 | ||
747 | TEXTS_FOR_ZONES("pgalloc") | |
748 | ||
749 | "pgfree", | |
750 | "pgactivate", | |
751 | "pgdeactivate", | |
752 | ||
753 | "pgfault", | |
754 | "pgmajfault", | |
755 | ||
756 | TEXTS_FOR_ZONES("pgrefill") | |
904249aa YH |
757 | TEXTS_FOR_ZONES("pgsteal_kswapd") |
758 | TEXTS_FOR_ZONES("pgsteal_direct") | |
fa25c503 KM |
759 | TEXTS_FOR_ZONES("pgscan_kswapd") |
760 | TEXTS_FOR_ZONES("pgscan_direct") | |
68243e76 | 761 | "pgscan_direct_throttle", |
fa25c503 KM |
762 | |
763 | #ifdef CONFIG_NUMA | |
764 | "zone_reclaim_failed", | |
765 | #endif | |
766 | "pginodesteal", | |
767 | "slabs_scanned", | |
fa25c503 KM |
768 | "kswapd_inodesteal", |
769 | "kswapd_low_wmark_hit_quickly", | |
770 | "kswapd_high_wmark_hit_quickly", | |
fa25c503 KM |
771 | "pageoutrun", |
772 | "allocstall", | |
773 | ||
774 | "pgrotated", | |
775 | ||
03c5a6e1 MG |
776 | #ifdef CONFIG_NUMA_BALANCING |
777 | "numa_pte_updates", | |
778 | "numa_hint_faults", | |
779 | "numa_hint_faults_local", | |
780 | "numa_pages_migrated", | |
781 | #endif | |
5647bc29 MG |
782 | #ifdef CONFIG_MIGRATION |
783 | "pgmigrate_success", | |
784 | "pgmigrate_fail", | |
785 | #endif | |
fa25c503 | 786 | #ifdef CONFIG_COMPACTION |
397487db MG |
787 | "compact_migrate_scanned", |
788 | "compact_free_scanned", | |
789 | "compact_isolated", | |
fa25c503 KM |
790 | "compact_stall", |
791 | "compact_fail", | |
792 | "compact_success", | |
793 | #endif | |
794 | ||
795 | #ifdef CONFIG_HUGETLB_PAGE | |
796 | "htlb_buddy_alloc_success", | |
797 | "htlb_buddy_alloc_fail", | |
798 | #endif | |
799 | "unevictable_pgs_culled", | |
800 | "unevictable_pgs_scanned", | |
801 | "unevictable_pgs_rescued", | |
802 | "unevictable_pgs_mlocked", | |
803 | "unevictable_pgs_munlocked", | |
804 | "unevictable_pgs_cleared", | |
805 | "unevictable_pgs_stranded", | |
fa25c503 KM |
806 | |
807 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
808 | "thp_fault_alloc", | |
809 | "thp_fault_fallback", | |
810 | "thp_collapse_alloc", | |
811 | "thp_collapse_alloc_failed", | |
812 | "thp_split", | |
d8a8e1f0 KS |
813 | "thp_zero_page_alloc", |
814 | "thp_zero_page_alloc_failed", | |
fa25c503 KM |
815 | #endif |
816 | ||
817 | #endif /* CONFIG_VM_EVENTS_COUNTERS */ | |
818 | }; | |
0d6617c7 | 819 | #endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA */ |
fa25c503 KM |
820 | |
821 | ||
d7a5752c | 822 | #ifdef CONFIG_PROC_FS |
467c996c MG |
823 | static void frag_show_print(struct seq_file *m, pg_data_t *pgdat, |
824 | struct zone *zone) | |
825 | { | |
826 | int order; | |
827 | ||
828 | seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name); | |
829 | for (order = 0; order < MAX_ORDER; ++order) | |
830 | seq_printf(m, "%6lu ", zone->free_area[order].nr_free); | |
831 | seq_putc(m, '\n'); | |
832 | } | |
833 | ||
834 | /* | |
835 | * This walks the free areas for each zone. | |
836 | */ | |
837 | static int frag_show(struct seq_file *m, void *arg) | |
838 | { | |
839 | pg_data_t *pgdat = (pg_data_t *)arg; | |
840 | walk_zones_in_node(m, pgdat, frag_show_print); | |
841 | return 0; | |
842 | } | |
843 | ||
844 | static void pagetypeinfo_showfree_print(struct seq_file *m, | |
845 | pg_data_t *pgdat, struct zone *zone) | |
846 | { | |
847 | int order, mtype; | |
848 | ||
849 | for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) { | |
850 | seq_printf(m, "Node %4d, zone %8s, type %12s ", | |
851 | pgdat->node_id, | |
852 | zone->name, | |
853 | migratetype_names[mtype]); | |
854 | for (order = 0; order < MAX_ORDER; ++order) { | |
855 | unsigned long freecount = 0; | |
856 | struct free_area *area; | |
857 | struct list_head *curr; | |
858 | ||
859 | area = &(zone->free_area[order]); | |
860 | ||
861 | list_for_each(curr, &area->free_list[mtype]) | |
862 | freecount++; | |
863 | seq_printf(m, "%6lu ", freecount); | |
864 | } | |
f6ac2354 CL |
865 | seq_putc(m, '\n'); |
866 | } | |
467c996c MG |
867 | } |
868 | ||
869 | /* Print out the free pages at each order for each migatetype */ | |
870 | static int pagetypeinfo_showfree(struct seq_file *m, void *arg) | |
871 | { | |
872 | int order; | |
873 | pg_data_t *pgdat = (pg_data_t *)arg; | |
874 | ||
875 | /* Print header */ | |
876 | seq_printf(m, "%-43s ", "Free pages count per migrate type at order"); | |
877 | for (order = 0; order < MAX_ORDER; ++order) | |
878 | seq_printf(m, "%6d ", order); | |
879 | seq_putc(m, '\n'); | |
880 | ||
881 | walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print); | |
882 | ||
883 | return 0; | |
884 | } | |
885 | ||
886 | static void pagetypeinfo_showblockcount_print(struct seq_file *m, | |
887 | pg_data_t *pgdat, struct zone *zone) | |
888 | { | |
889 | int mtype; | |
890 | unsigned long pfn; | |
891 | unsigned long start_pfn = zone->zone_start_pfn; | |
108bcc96 | 892 | unsigned long end_pfn = zone_end_pfn(zone); |
467c996c MG |
893 | unsigned long count[MIGRATE_TYPES] = { 0, }; |
894 | ||
895 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { | |
896 | struct page *page; | |
897 | ||
898 | if (!pfn_valid(pfn)) | |
899 | continue; | |
900 | ||
901 | page = pfn_to_page(pfn); | |
eb33575c MG |
902 | |
903 | /* Watch for unexpected holes punched in the memmap */ | |
904 | if (!memmap_valid_within(pfn, page, zone)) | |
e80d6a24 | 905 | continue; |
eb33575c | 906 | |
467c996c MG |
907 | mtype = get_pageblock_migratetype(page); |
908 | ||
e80d6a24 MG |
909 | if (mtype < MIGRATE_TYPES) |
910 | count[mtype]++; | |
467c996c MG |
911 | } |
912 | ||
913 | /* Print counts */ | |
914 | seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name); | |
915 | for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) | |
916 | seq_printf(m, "%12lu ", count[mtype]); | |
917 | seq_putc(m, '\n'); | |
918 | } | |
919 | ||
920 | /* Print out the free pages at each order for each migratetype */ | |
921 | static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg) | |
922 | { | |
923 | int mtype; | |
924 | pg_data_t *pgdat = (pg_data_t *)arg; | |
925 | ||
926 | seq_printf(m, "\n%-23s", "Number of blocks type "); | |
927 | for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) | |
928 | seq_printf(m, "%12s ", migratetype_names[mtype]); | |
929 | seq_putc(m, '\n'); | |
930 | walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print); | |
931 | ||
932 | return 0; | |
933 | } | |
934 | ||
935 | /* | |
936 | * This prints out statistics in relation to grouping pages by mobility. | |
937 | * It is expensive to collect so do not constantly read the file. | |
938 | */ | |
939 | static int pagetypeinfo_show(struct seq_file *m, void *arg) | |
940 | { | |
941 | pg_data_t *pgdat = (pg_data_t *)arg; | |
942 | ||
41b25a37 | 943 | /* check memoryless node */ |
a47b53c5 | 944 | if (!node_state(pgdat->node_id, N_MEMORY)) |
41b25a37 KM |
945 | return 0; |
946 | ||
467c996c MG |
947 | seq_printf(m, "Page block order: %d\n", pageblock_order); |
948 | seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages); | |
949 | seq_putc(m, '\n'); | |
950 | pagetypeinfo_showfree(m, pgdat); | |
951 | pagetypeinfo_showblockcount(m, pgdat); | |
952 | ||
f6ac2354 CL |
953 | return 0; |
954 | } | |
955 | ||
8f32f7e5 | 956 | static const struct seq_operations fragmentation_op = { |
f6ac2354 CL |
957 | .start = frag_start, |
958 | .next = frag_next, | |
959 | .stop = frag_stop, | |
960 | .show = frag_show, | |
961 | }; | |
962 | ||
8f32f7e5 AD |
963 | static int fragmentation_open(struct inode *inode, struct file *file) |
964 | { | |
965 | return seq_open(file, &fragmentation_op); | |
966 | } | |
967 | ||
968 | static const struct file_operations fragmentation_file_operations = { | |
969 | .open = fragmentation_open, | |
970 | .read = seq_read, | |
971 | .llseek = seq_lseek, | |
972 | .release = seq_release, | |
973 | }; | |
974 | ||
74e2e8e8 | 975 | static const struct seq_operations pagetypeinfo_op = { |
467c996c MG |
976 | .start = frag_start, |
977 | .next = frag_next, | |
978 | .stop = frag_stop, | |
979 | .show = pagetypeinfo_show, | |
980 | }; | |
981 | ||
74e2e8e8 AD |
982 | static int pagetypeinfo_open(struct inode *inode, struct file *file) |
983 | { | |
984 | return seq_open(file, &pagetypeinfo_op); | |
985 | } | |
986 | ||
987 | static const struct file_operations pagetypeinfo_file_ops = { | |
988 | .open = pagetypeinfo_open, | |
989 | .read = seq_read, | |
990 | .llseek = seq_lseek, | |
991 | .release = seq_release, | |
992 | }; | |
993 | ||
467c996c MG |
994 | static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat, |
995 | struct zone *zone) | |
f6ac2354 | 996 | { |
467c996c MG |
997 | int i; |
998 | seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name); | |
999 | seq_printf(m, | |
1000 | "\n pages free %lu" | |
1001 | "\n min %lu" | |
1002 | "\n low %lu" | |
1003 | "\n high %lu" | |
08d9ae7c | 1004 | "\n scanned %lu" |
467c996c | 1005 | "\n spanned %lu" |
9feedc9d JL |
1006 | "\n present %lu" |
1007 | "\n managed %lu", | |
88f5acf8 | 1008 | zone_page_state(zone, NR_FREE_PAGES), |
41858966 MG |
1009 | min_wmark_pages(zone), |
1010 | low_wmark_pages(zone), | |
1011 | high_wmark_pages(zone), | |
467c996c | 1012 | zone->pages_scanned, |
467c996c | 1013 | zone->spanned_pages, |
9feedc9d JL |
1014 | zone->present_pages, |
1015 | zone->managed_pages); | |
467c996c MG |
1016 | |
1017 | for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) | |
1018 | seq_printf(m, "\n %-12s %lu", vmstat_text[i], | |
1019 | zone_page_state(zone, i)); | |
1020 | ||
1021 | seq_printf(m, | |
1022 | "\n protection: (%lu", | |
1023 | zone->lowmem_reserve[0]); | |
1024 | for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++) | |
1025 | seq_printf(m, ", %lu", zone->lowmem_reserve[i]); | |
1026 | seq_printf(m, | |
1027 | ")" | |
1028 | "\n pagesets"); | |
1029 | for_each_online_cpu(i) { | |
1030 | struct per_cpu_pageset *pageset; | |
467c996c | 1031 | |
99dcc3e5 | 1032 | pageset = per_cpu_ptr(zone->pageset, i); |
3dfa5721 CL |
1033 | seq_printf(m, |
1034 | "\n cpu: %i" | |
1035 | "\n count: %i" | |
1036 | "\n high: %i" | |
1037 | "\n batch: %i", | |
1038 | i, | |
1039 | pageset->pcp.count, | |
1040 | pageset->pcp.high, | |
1041 | pageset->pcp.batch); | |
df9ecaba | 1042 | #ifdef CONFIG_SMP |
467c996c MG |
1043 | seq_printf(m, "\n vm stats threshold: %d", |
1044 | pageset->stat_threshold); | |
df9ecaba | 1045 | #endif |
f6ac2354 | 1046 | } |
467c996c MG |
1047 | seq_printf(m, |
1048 | "\n all_unreclaimable: %u" | |
556adecb RR |
1049 | "\n start_pfn: %lu" |
1050 | "\n inactive_ratio: %u", | |
93e4a89a | 1051 | zone->all_unreclaimable, |
556adecb RR |
1052 | zone->zone_start_pfn, |
1053 | zone->inactive_ratio); | |
467c996c MG |
1054 | seq_putc(m, '\n'); |
1055 | } | |
1056 | ||
1057 | /* | |
1058 | * Output information about zones in @pgdat. | |
1059 | */ | |
1060 | static int zoneinfo_show(struct seq_file *m, void *arg) | |
1061 | { | |
1062 | pg_data_t *pgdat = (pg_data_t *)arg; | |
1063 | walk_zones_in_node(m, pgdat, zoneinfo_show_print); | |
f6ac2354 CL |
1064 | return 0; |
1065 | } | |
1066 | ||
5c9fe628 | 1067 | static const struct seq_operations zoneinfo_op = { |
f6ac2354 CL |
1068 | .start = frag_start, /* iterate over all zones. The same as in |
1069 | * fragmentation. */ | |
1070 | .next = frag_next, | |
1071 | .stop = frag_stop, | |
1072 | .show = zoneinfo_show, | |
1073 | }; | |
1074 | ||
5c9fe628 AD |
1075 | static int zoneinfo_open(struct inode *inode, struct file *file) |
1076 | { | |
1077 | return seq_open(file, &zoneinfo_op); | |
1078 | } | |
1079 | ||
1080 | static const struct file_operations proc_zoneinfo_file_operations = { | |
1081 | .open = zoneinfo_open, | |
1082 | .read = seq_read, | |
1083 | .llseek = seq_lseek, | |
1084 | .release = seq_release, | |
1085 | }; | |
1086 | ||
79da826a MR |
1087 | enum writeback_stat_item { |
1088 | NR_DIRTY_THRESHOLD, | |
1089 | NR_DIRTY_BG_THRESHOLD, | |
1090 | NR_VM_WRITEBACK_STAT_ITEMS, | |
1091 | }; | |
1092 | ||
f6ac2354 CL |
1093 | static void *vmstat_start(struct seq_file *m, loff_t *pos) |
1094 | { | |
2244b95a | 1095 | unsigned long *v; |
79da826a | 1096 | int i, stat_items_size; |
f6ac2354 CL |
1097 | |
1098 | if (*pos >= ARRAY_SIZE(vmstat_text)) | |
1099 | return NULL; | |
79da826a MR |
1100 | stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) + |
1101 | NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long); | |
f6ac2354 | 1102 | |
f8891e5e | 1103 | #ifdef CONFIG_VM_EVENT_COUNTERS |
79da826a | 1104 | stat_items_size += sizeof(struct vm_event_state); |
f8891e5e | 1105 | #endif |
79da826a MR |
1106 | |
1107 | v = kmalloc(stat_items_size, GFP_KERNEL); | |
2244b95a CL |
1108 | m->private = v; |
1109 | if (!v) | |
f6ac2354 | 1110 | return ERR_PTR(-ENOMEM); |
2244b95a CL |
1111 | for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) |
1112 | v[i] = global_page_state(i); | |
79da826a MR |
1113 | v += NR_VM_ZONE_STAT_ITEMS; |
1114 | ||
1115 | global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD, | |
1116 | v + NR_DIRTY_THRESHOLD); | |
1117 | v += NR_VM_WRITEBACK_STAT_ITEMS; | |
1118 | ||
f8891e5e | 1119 | #ifdef CONFIG_VM_EVENT_COUNTERS |
79da826a MR |
1120 | all_vm_events(v); |
1121 | v[PGPGIN] /= 2; /* sectors -> kbytes */ | |
1122 | v[PGPGOUT] /= 2; | |
f8891e5e | 1123 | #endif |
ff8b16d7 | 1124 | return (unsigned long *)m->private + *pos; |
f6ac2354 CL |
1125 | } |
1126 | ||
1127 | static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos) | |
1128 | { | |
1129 | (*pos)++; | |
1130 | if (*pos >= ARRAY_SIZE(vmstat_text)) | |
1131 | return NULL; | |
1132 | return (unsigned long *)m->private + *pos; | |
1133 | } | |
1134 | ||
1135 | static int vmstat_show(struct seq_file *m, void *arg) | |
1136 | { | |
1137 | unsigned long *l = arg; | |
1138 | unsigned long off = l - (unsigned long *)m->private; | |
1139 | ||
1140 | seq_printf(m, "%s %lu\n", vmstat_text[off], *l); | |
1141 | return 0; | |
1142 | } | |
1143 | ||
1144 | static void vmstat_stop(struct seq_file *m, void *arg) | |
1145 | { | |
1146 | kfree(m->private); | |
1147 | m->private = NULL; | |
1148 | } | |
1149 | ||
b6aa44ab | 1150 | static const struct seq_operations vmstat_op = { |
f6ac2354 CL |
1151 | .start = vmstat_start, |
1152 | .next = vmstat_next, | |
1153 | .stop = vmstat_stop, | |
1154 | .show = vmstat_show, | |
1155 | }; | |
1156 | ||
b6aa44ab AD |
1157 | static int vmstat_open(struct inode *inode, struct file *file) |
1158 | { | |
1159 | return seq_open(file, &vmstat_op); | |
1160 | } | |
1161 | ||
1162 | static const struct file_operations proc_vmstat_file_operations = { | |
1163 | .open = vmstat_open, | |
1164 | .read = seq_read, | |
1165 | .llseek = seq_lseek, | |
1166 | .release = seq_release, | |
1167 | }; | |
f6ac2354 CL |
1168 | #endif /* CONFIG_PROC_FS */ |
1169 | ||
df9ecaba | 1170 | #ifdef CONFIG_SMP |
d1187ed2 | 1171 | static DEFINE_PER_CPU(struct delayed_work, vmstat_work); |
77461ab3 | 1172 | int sysctl_stat_interval __read_mostly = HZ; |
d1187ed2 CL |
1173 | |
1174 | static void vmstat_update(struct work_struct *w) | |
1175 | { | |
1176 | refresh_cpu_vm_stats(smp_processor_id()); | |
77461ab3 | 1177 | schedule_delayed_work(&__get_cpu_var(vmstat_work), |
98f4ebb2 | 1178 | round_jiffies_relative(sysctl_stat_interval)); |
d1187ed2 CL |
1179 | } |
1180 | ||
42614fcd | 1181 | static void __cpuinit start_cpu_timer(int cpu) |
d1187ed2 | 1182 | { |
1871e52c | 1183 | struct delayed_work *work = &per_cpu(vmstat_work, cpu); |
d1187ed2 | 1184 | |
203b42f7 | 1185 | INIT_DEFERRABLE_WORK(work, vmstat_update); |
1871e52c | 1186 | schedule_delayed_work_on(cpu, work, __round_jiffies_relative(HZ, cpu)); |
d1187ed2 CL |
1187 | } |
1188 | ||
df9ecaba CL |
1189 | /* |
1190 | * Use the cpu notifier to insure that the thresholds are recalculated | |
1191 | * when necessary. | |
1192 | */ | |
1193 | static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb, | |
1194 | unsigned long action, | |
1195 | void *hcpu) | |
1196 | { | |
d1187ed2 CL |
1197 | long cpu = (long)hcpu; |
1198 | ||
df9ecaba | 1199 | switch (action) { |
d1187ed2 CL |
1200 | case CPU_ONLINE: |
1201 | case CPU_ONLINE_FROZEN: | |
5ee28a44 | 1202 | refresh_zone_stat_thresholds(); |
d1187ed2 | 1203 | start_cpu_timer(cpu); |
ad596925 | 1204 | node_set_state(cpu_to_node(cpu), N_CPU); |
d1187ed2 CL |
1205 | break; |
1206 | case CPU_DOWN_PREPARE: | |
1207 | case CPU_DOWN_PREPARE_FROZEN: | |
afe2c511 | 1208 | cancel_delayed_work_sync(&per_cpu(vmstat_work, cpu)); |
d1187ed2 CL |
1209 | per_cpu(vmstat_work, cpu).work.func = NULL; |
1210 | break; | |
1211 | case CPU_DOWN_FAILED: | |
1212 | case CPU_DOWN_FAILED_FROZEN: | |
1213 | start_cpu_timer(cpu); | |
1214 | break; | |
ce421c79 | 1215 | case CPU_DEAD: |
8bb78442 | 1216 | case CPU_DEAD_FROZEN: |
ce421c79 AW |
1217 | refresh_zone_stat_thresholds(); |
1218 | break; | |
1219 | default: | |
1220 | break; | |
df9ecaba CL |
1221 | } |
1222 | return NOTIFY_OK; | |
1223 | } | |
1224 | ||
1225 | static struct notifier_block __cpuinitdata vmstat_notifier = | |
1226 | { &vmstat_cpuup_callback, NULL, 0 }; | |
8f32f7e5 | 1227 | #endif |
df9ecaba | 1228 | |
e2fc88d0 | 1229 | static int __init setup_vmstat(void) |
df9ecaba | 1230 | { |
8f32f7e5 | 1231 | #ifdef CONFIG_SMP |
d1187ed2 CL |
1232 | int cpu; |
1233 | ||
df9ecaba | 1234 | register_cpu_notifier(&vmstat_notifier); |
d1187ed2 CL |
1235 | |
1236 | for_each_online_cpu(cpu) | |
1237 | start_cpu_timer(cpu); | |
8f32f7e5 AD |
1238 | #endif |
1239 | #ifdef CONFIG_PROC_FS | |
1240 | proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations); | |
74e2e8e8 | 1241 | proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops); |
b6aa44ab | 1242 | proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations); |
5c9fe628 | 1243 | proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations); |
8f32f7e5 | 1244 | #endif |
df9ecaba CL |
1245 | return 0; |
1246 | } | |
1247 | module_init(setup_vmstat) | |
d7a5752c MG |
1248 | |
1249 | #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION) | |
1250 | #include <linux/debugfs.h> | |
1251 | ||
d7a5752c MG |
1252 | |
1253 | /* | |
1254 | * Return an index indicating how much of the available free memory is | |
1255 | * unusable for an allocation of the requested size. | |
1256 | */ | |
1257 | static int unusable_free_index(unsigned int order, | |
1258 | struct contig_page_info *info) | |
1259 | { | |
1260 | /* No free memory is interpreted as all free memory is unusable */ | |
1261 | if (info->free_pages == 0) | |
1262 | return 1000; | |
1263 | ||
1264 | /* | |
1265 | * Index should be a value between 0 and 1. Return a value to 3 | |
1266 | * decimal places. | |
1267 | * | |
1268 | * 0 => no fragmentation | |
1269 | * 1 => high fragmentation | |
1270 | */ | |
1271 | return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages); | |
1272 | ||
1273 | } | |
1274 | ||
1275 | static void unusable_show_print(struct seq_file *m, | |
1276 | pg_data_t *pgdat, struct zone *zone) | |
1277 | { | |
1278 | unsigned int order; | |
1279 | int index; | |
1280 | struct contig_page_info info; | |
1281 | ||
1282 | seq_printf(m, "Node %d, zone %8s ", | |
1283 | pgdat->node_id, | |
1284 | zone->name); | |
1285 | for (order = 0; order < MAX_ORDER; ++order) { | |
1286 | fill_contig_page_info(zone, order, &info); | |
1287 | index = unusable_free_index(order, &info); | |
1288 | seq_printf(m, "%d.%03d ", index / 1000, index % 1000); | |
1289 | } | |
1290 | ||
1291 | seq_putc(m, '\n'); | |
1292 | } | |
1293 | ||
1294 | /* | |
1295 | * Display unusable free space index | |
1296 | * | |
1297 | * The unusable free space index measures how much of the available free | |
1298 | * memory cannot be used to satisfy an allocation of a given size and is a | |
1299 | * value between 0 and 1. The higher the value, the more of free memory is | |
1300 | * unusable and by implication, the worse the external fragmentation is. This | |
1301 | * can be expressed as a percentage by multiplying by 100. | |
1302 | */ | |
1303 | static int unusable_show(struct seq_file *m, void *arg) | |
1304 | { | |
1305 | pg_data_t *pgdat = (pg_data_t *)arg; | |
1306 | ||
1307 | /* check memoryless node */ | |
a47b53c5 | 1308 | if (!node_state(pgdat->node_id, N_MEMORY)) |
d7a5752c MG |
1309 | return 0; |
1310 | ||
1311 | walk_zones_in_node(m, pgdat, unusable_show_print); | |
1312 | ||
1313 | return 0; | |
1314 | } | |
1315 | ||
1316 | static const struct seq_operations unusable_op = { | |
1317 | .start = frag_start, | |
1318 | .next = frag_next, | |
1319 | .stop = frag_stop, | |
1320 | .show = unusable_show, | |
1321 | }; | |
1322 | ||
1323 | static int unusable_open(struct inode *inode, struct file *file) | |
1324 | { | |
1325 | return seq_open(file, &unusable_op); | |
1326 | } | |
1327 | ||
1328 | static const struct file_operations unusable_file_ops = { | |
1329 | .open = unusable_open, | |
1330 | .read = seq_read, | |
1331 | .llseek = seq_lseek, | |
1332 | .release = seq_release, | |
1333 | }; | |
1334 | ||
f1a5ab12 MG |
1335 | static void extfrag_show_print(struct seq_file *m, |
1336 | pg_data_t *pgdat, struct zone *zone) | |
1337 | { | |
1338 | unsigned int order; | |
1339 | int index; | |
1340 | ||
1341 | /* Alloc on stack as interrupts are disabled for zone walk */ | |
1342 | struct contig_page_info info; | |
1343 | ||
1344 | seq_printf(m, "Node %d, zone %8s ", | |
1345 | pgdat->node_id, | |
1346 | zone->name); | |
1347 | for (order = 0; order < MAX_ORDER; ++order) { | |
1348 | fill_contig_page_info(zone, order, &info); | |
56de7263 | 1349 | index = __fragmentation_index(order, &info); |
f1a5ab12 MG |
1350 | seq_printf(m, "%d.%03d ", index / 1000, index % 1000); |
1351 | } | |
1352 | ||
1353 | seq_putc(m, '\n'); | |
1354 | } | |
1355 | ||
1356 | /* | |
1357 | * Display fragmentation index for orders that allocations would fail for | |
1358 | */ | |
1359 | static int extfrag_show(struct seq_file *m, void *arg) | |
1360 | { | |
1361 | pg_data_t *pgdat = (pg_data_t *)arg; | |
1362 | ||
1363 | walk_zones_in_node(m, pgdat, extfrag_show_print); | |
1364 | ||
1365 | return 0; | |
1366 | } | |
1367 | ||
1368 | static const struct seq_operations extfrag_op = { | |
1369 | .start = frag_start, | |
1370 | .next = frag_next, | |
1371 | .stop = frag_stop, | |
1372 | .show = extfrag_show, | |
1373 | }; | |
1374 | ||
1375 | static int extfrag_open(struct inode *inode, struct file *file) | |
1376 | { | |
1377 | return seq_open(file, &extfrag_op); | |
1378 | } | |
1379 | ||
1380 | static const struct file_operations extfrag_file_ops = { | |
1381 | .open = extfrag_open, | |
1382 | .read = seq_read, | |
1383 | .llseek = seq_lseek, | |
1384 | .release = seq_release, | |
1385 | }; | |
1386 | ||
d7a5752c MG |
1387 | static int __init extfrag_debug_init(void) |
1388 | { | |
bde8bd8a S |
1389 | struct dentry *extfrag_debug_root; |
1390 | ||
d7a5752c MG |
1391 | extfrag_debug_root = debugfs_create_dir("extfrag", NULL); |
1392 | if (!extfrag_debug_root) | |
1393 | return -ENOMEM; | |
1394 | ||
1395 | if (!debugfs_create_file("unusable_index", 0444, | |
1396 | extfrag_debug_root, NULL, &unusable_file_ops)) | |
bde8bd8a | 1397 | goto fail; |
d7a5752c | 1398 | |
f1a5ab12 MG |
1399 | if (!debugfs_create_file("extfrag_index", 0444, |
1400 | extfrag_debug_root, NULL, &extfrag_file_ops)) | |
bde8bd8a | 1401 | goto fail; |
f1a5ab12 | 1402 | |
d7a5752c | 1403 | return 0; |
bde8bd8a S |
1404 | fail: |
1405 | debugfs_remove_recursive(extfrag_debug_root); | |
1406 | return -ENOMEM; | |
d7a5752c MG |
1407 | } |
1408 | ||
1409 | module_init(extfrag_debug_init); | |
1410 | #endif |