]>
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> | |
7cc36bbd | 10 | * Copyright (C) 2008-2014 Christoph Lameter |
f6ac2354 | 11 | */ |
8f32f7e5 | 12 | #include <linux/fs.h> |
f6ac2354 | 13 | #include <linux/mm.h> |
4e950f6f | 14 | #include <linux/err.h> |
2244b95a | 15 | #include <linux/module.h> |
5a0e3ad6 | 16 | #include <linux/slab.h> |
df9ecaba | 17 | #include <linux/cpu.h> |
7cc36bbd | 18 | #include <linux/cpumask.h> |
c748e134 | 19 | #include <linux/vmstat.h> |
3c486871 AM |
20 | #include <linux/proc_fs.h> |
21 | #include <linux/seq_file.h> | |
22 | #include <linux/debugfs.h> | |
e8edc6e0 | 23 | #include <linux/sched.h> |
f1a5ab12 | 24 | #include <linux/math64.h> |
79da826a | 25 | #include <linux/writeback.h> |
36deb0be | 26 | #include <linux/compaction.h> |
6e543d57 | 27 | #include <linux/mm_inline.h> |
48c96a36 JK |
28 | #include <linux/page_ext.h> |
29 | #include <linux/page_owner.h> | |
6e543d57 LD |
30 | |
31 | #include "internal.h" | |
f6ac2354 | 32 | |
f8891e5e CL |
33 | #ifdef CONFIG_VM_EVENT_COUNTERS |
34 | DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}}; | |
35 | EXPORT_PER_CPU_SYMBOL(vm_event_states); | |
36 | ||
31f961a8 | 37 | static void sum_vm_events(unsigned long *ret) |
f8891e5e | 38 | { |
9eccf2a8 | 39 | int cpu; |
f8891e5e CL |
40 | int i; |
41 | ||
42 | memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long)); | |
43 | ||
31f961a8 | 44 | for_each_online_cpu(cpu) { |
f8891e5e CL |
45 | struct vm_event_state *this = &per_cpu(vm_event_states, cpu); |
46 | ||
f8891e5e CL |
47 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) |
48 | ret[i] += this->event[i]; | |
49 | } | |
50 | } | |
51 | ||
52 | /* | |
53 | * Accumulate the vm event counters across all CPUs. | |
54 | * The result is unavoidably approximate - it can change | |
55 | * during and after execution of this function. | |
56 | */ | |
57 | void all_vm_events(unsigned long *ret) | |
58 | { | |
b5be1132 | 59 | get_online_cpus(); |
31f961a8 | 60 | sum_vm_events(ret); |
b5be1132 | 61 | put_online_cpus(); |
f8891e5e | 62 | } |
32dd66fc | 63 | EXPORT_SYMBOL_GPL(all_vm_events); |
f8891e5e | 64 | |
f8891e5e CL |
65 | /* |
66 | * Fold the foreign cpu events into our own. | |
67 | * | |
68 | * This is adding to the events on one processor | |
69 | * but keeps the global counts constant. | |
70 | */ | |
71 | void vm_events_fold_cpu(int cpu) | |
72 | { | |
73 | struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu); | |
74 | int i; | |
75 | ||
76 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { | |
77 | count_vm_events(i, fold_state->event[i]); | |
78 | fold_state->event[i] = 0; | |
79 | } | |
80 | } | |
f8891e5e CL |
81 | |
82 | #endif /* CONFIG_VM_EVENT_COUNTERS */ | |
83 | ||
2244b95a CL |
84 | /* |
85 | * Manage combined zone based / global counters | |
86 | * | |
87 | * vm_stat contains the global counters | |
88 | */ | |
75ef7184 | 89 | atomic_long_t vm_zone_stat[NR_VM_ZONE_STAT_ITEMS] __cacheline_aligned_in_smp; |
3a321d2a | 90 | atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS] __cacheline_aligned_in_smp; |
75ef7184 MG |
91 | atomic_long_t vm_node_stat[NR_VM_NODE_STAT_ITEMS] __cacheline_aligned_in_smp; |
92 | EXPORT_SYMBOL(vm_zone_stat); | |
3a321d2a | 93 | EXPORT_SYMBOL(vm_numa_stat); |
75ef7184 | 94 | EXPORT_SYMBOL(vm_node_stat); |
2244b95a CL |
95 | |
96 | #ifdef CONFIG_SMP | |
97 | ||
b44129b3 | 98 | int calculate_pressure_threshold(struct zone *zone) |
88f5acf8 MG |
99 | { |
100 | int threshold; | |
101 | int watermark_distance; | |
102 | ||
103 | /* | |
104 | * As vmstats are not up to date, there is drift between the estimated | |
105 | * and real values. For high thresholds and a high number of CPUs, it | |
106 | * is possible for the min watermark to be breached while the estimated | |
107 | * value looks fine. The pressure threshold is a reduced value such | |
108 | * that even the maximum amount of drift will not accidentally breach | |
109 | * the min watermark | |
110 | */ | |
111 | watermark_distance = low_wmark_pages(zone) - min_wmark_pages(zone); | |
112 | threshold = max(1, (int)(watermark_distance / num_online_cpus())); | |
113 | ||
114 | /* | |
115 | * Maximum threshold is 125 | |
116 | */ | |
117 | threshold = min(125, threshold); | |
118 | ||
119 | return threshold; | |
120 | } | |
121 | ||
b44129b3 | 122 | int calculate_normal_threshold(struct zone *zone) |
df9ecaba CL |
123 | { |
124 | int threshold; | |
125 | int mem; /* memory in 128 MB units */ | |
126 | ||
127 | /* | |
128 | * The threshold scales with the number of processors and the amount | |
129 | * of memory per zone. More memory means that we can defer updates for | |
130 | * longer, more processors could lead to more contention. | |
131 | * fls() is used to have a cheap way of logarithmic scaling. | |
132 | * | |
133 | * Some sample thresholds: | |
134 | * | |
135 | * Threshold Processors (fls) Zonesize fls(mem+1) | |
136 | * ------------------------------------------------------------------ | |
137 | * 8 1 1 0.9-1 GB 4 | |
138 | * 16 2 2 0.9-1 GB 4 | |
139 | * 20 2 2 1-2 GB 5 | |
140 | * 24 2 2 2-4 GB 6 | |
141 | * 28 2 2 4-8 GB 7 | |
142 | * 32 2 2 8-16 GB 8 | |
143 | * 4 2 2 <128M 1 | |
144 | * 30 4 3 2-4 GB 5 | |
145 | * 48 4 3 8-16 GB 8 | |
146 | * 32 8 4 1-2 GB 4 | |
147 | * 32 8 4 0.9-1GB 4 | |
148 | * 10 16 5 <128M 1 | |
149 | * 40 16 5 900M 4 | |
150 | * 70 64 7 2-4 GB 5 | |
151 | * 84 64 7 4-8 GB 6 | |
152 | * 108 512 9 4-8 GB 6 | |
153 | * 125 1024 10 8-16 GB 8 | |
154 | * 125 1024 10 16-32 GB 9 | |
155 | */ | |
156 | ||
b40da049 | 157 | mem = zone->managed_pages >> (27 - PAGE_SHIFT); |
df9ecaba CL |
158 | |
159 | threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem)); | |
160 | ||
161 | /* | |
162 | * Maximum threshold is 125 | |
163 | */ | |
164 | threshold = min(125, threshold); | |
165 | ||
166 | return threshold; | |
167 | } | |
2244b95a CL |
168 | |
169 | /* | |
df9ecaba | 170 | * Refresh the thresholds for each zone. |
2244b95a | 171 | */ |
a6cccdc3 | 172 | void refresh_zone_stat_thresholds(void) |
2244b95a | 173 | { |
75ef7184 | 174 | struct pglist_data *pgdat; |
df9ecaba CL |
175 | struct zone *zone; |
176 | int cpu; | |
177 | int threshold; | |
178 | ||
75ef7184 MG |
179 | /* Zero current pgdat thresholds */ |
180 | for_each_online_pgdat(pgdat) { | |
181 | for_each_online_cpu(cpu) { | |
182 | per_cpu_ptr(pgdat->per_cpu_nodestats, cpu)->stat_threshold = 0; | |
183 | } | |
184 | } | |
185 | ||
ee99c71c | 186 | for_each_populated_zone(zone) { |
75ef7184 | 187 | struct pglist_data *pgdat = zone->zone_pgdat; |
aa454840 CL |
188 | unsigned long max_drift, tolerate_drift; |
189 | ||
b44129b3 | 190 | threshold = calculate_normal_threshold(zone); |
df9ecaba | 191 | |
75ef7184 MG |
192 | for_each_online_cpu(cpu) { |
193 | int pgdat_threshold; | |
194 | ||
99dcc3e5 CL |
195 | per_cpu_ptr(zone->pageset, cpu)->stat_threshold |
196 | = threshold; | |
3a321d2a KW |
197 | #ifdef CONFIG_NUMA |
198 | per_cpu_ptr(zone->pageset, cpu)->numa_stat_threshold | |
199 | = threshold; | |
200 | #endif | |
75ef7184 MG |
201 | /* Base nodestat threshold on the largest populated zone. */ |
202 | pgdat_threshold = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu)->stat_threshold; | |
203 | per_cpu_ptr(pgdat->per_cpu_nodestats, cpu)->stat_threshold | |
204 | = max(threshold, pgdat_threshold); | |
205 | } | |
206 | ||
aa454840 CL |
207 | /* |
208 | * Only set percpu_drift_mark if there is a danger that | |
209 | * NR_FREE_PAGES reports the low watermark is ok when in fact | |
210 | * the min watermark could be breached by an allocation | |
211 | */ | |
212 | tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone); | |
213 | max_drift = num_online_cpus() * threshold; | |
214 | if (max_drift > tolerate_drift) | |
215 | zone->percpu_drift_mark = high_wmark_pages(zone) + | |
216 | max_drift; | |
df9ecaba | 217 | } |
2244b95a CL |
218 | } |
219 | ||
b44129b3 MG |
220 | void set_pgdat_percpu_threshold(pg_data_t *pgdat, |
221 | int (*calculate_pressure)(struct zone *)) | |
88f5acf8 MG |
222 | { |
223 | struct zone *zone; | |
224 | int cpu; | |
225 | int threshold; | |
226 | int i; | |
227 | ||
88f5acf8 MG |
228 | for (i = 0; i < pgdat->nr_zones; i++) { |
229 | zone = &pgdat->node_zones[i]; | |
230 | if (!zone->percpu_drift_mark) | |
231 | continue; | |
232 | ||
b44129b3 | 233 | threshold = (*calculate_pressure)(zone); |
3a321d2a | 234 | for_each_online_cpu(cpu) { |
88f5acf8 MG |
235 | per_cpu_ptr(zone->pageset, cpu)->stat_threshold |
236 | = threshold; | |
3a321d2a KW |
237 | #ifdef CONFIG_NUMA |
238 | per_cpu_ptr(zone->pageset, cpu)->numa_stat_threshold | |
239 | = threshold; | |
240 | #endif | |
241 | } | |
88f5acf8 | 242 | } |
88f5acf8 MG |
243 | } |
244 | ||
2244b95a | 245 | /* |
bea04b07 JZ |
246 | * For use when we know that interrupts are disabled, |
247 | * or when we know that preemption is disabled and that | |
248 | * particular counter cannot be updated from interrupt context. | |
2244b95a CL |
249 | */ |
250 | void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item, | |
6cdb18ad | 251 | long delta) |
2244b95a | 252 | { |
12938a92 CL |
253 | struct per_cpu_pageset __percpu *pcp = zone->pageset; |
254 | s8 __percpu *p = pcp->vm_stat_diff + item; | |
2244b95a | 255 | long x; |
12938a92 CL |
256 | long t; |
257 | ||
258 | x = delta + __this_cpu_read(*p); | |
2244b95a | 259 | |
12938a92 | 260 | t = __this_cpu_read(pcp->stat_threshold); |
2244b95a | 261 | |
12938a92 | 262 | if (unlikely(x > t || x < -t)) { |
2244b95a CL |
263 | zone_page_state_add(x, zone, item); |
264 | x = 0; | |
265 | } | |
12938a92 | 266 | __this_cpu_write(*p, x); |
2244b95a CL |
267 | } |
268 | EXPORT_SYMBOL(__mod_zone_page_state); | |
269 | ||
75ef7184 MG |
270 | void __mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item, |
271 | long delta) | |
272 | { | |
273 | struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats; | |
274 | s8 __percpu *p = pcp->vm_node_stat_diff + item; | |
275 | long x; | |
276 | long t; | |
277 | ||
278 | x = delta + __this_cpu_read(*p); | |
279 | ||
280 | t = __this_cpu_read(pcp->stat_threshold); | |
281 | ||
282 | if (unlikely(x > t || x < -t)) { | |
283 | node_page_state_add(x, pgdat, item); | |
284 | x = 0; | |
285 | } | |
286 | __this_cpu_write(*p, x); | |
287 | } | |
288 | EXPORT_SYMBOL(__mod_node_page_state); | |
289 | ||
2244b95a CL |
290 | /* |
291 | * Optimized increment and decrement functions. | |
292 | * | |
293 | * These are only for a single page and therefore can take a struct page * | |
294 | * argument instead of struct zone *. This allows the inclusion of the code | |
295 | * generated for page_zone(page) into the optimized functions. | |
296 | * | |
297 | * No overflow check is necessary and therefore the differential can be | |
298 | * incremented or decremented in place which may allow the compilers to | |
299 | * generate better code. | |
2244b95a CL |
300 | * The increment or decrement is known and therefore one boundary check can |
301 | * be omitted. | |
302 | * | |
df9ecaba CL |
303 | * NOTE: These functions are very performance sensitive. Change only |
304 | * with care. | |
305 | * | |
2244b95a CL |
306 | * Some processors have inc/dec instructions that are atomic vs an interrupt. |
307 | * However, the code must first determine the differential location in a zone | |
308 | * based on the processor number and then inc/dec the counter. There is no | |
309 | * guarantee without disabling preemption that the processor will not change | |
310 | * in between and therefore the atomicity vs. interrupt cannot be exploited | |
311 | * in a useful way here. | |
312 | */ | |
c8785385 | 313 | void __inc_zone_state(struct zone *zone, enum zone_stat_item item) |
2244b95a | 314 | { |
12938a92 CL |
315 | struct per_cpu_pageset __percpu *pcp = zone->pageset; |
316 | s8 __percpu *p = pcp->vm_stat_diff + item; | |
317 | s8 v, t; | |
2244b95a | 318 | |
908ee0f1 | 319 | v = __this_cpu_inc_return(*p); |
12938a92 CL |
320 | t = __this_cpu_read(pcp->stat_threshold); |
321 | if (unlikely(v > t)) { | |
322 | s8 overstep = t >> 1; | |
df9ecaba | 323 | |
12938a92 CL |
324 | zone_page_state_add(v + overstep, zone, item); |
325 | __this_cpu_write(*p, -overstep); | |
2244b95a CL |
326 | } |
327 | } | |
ca889e6c | 328 | |
75ef7184 MG |
329 | void __inc_node_state(struct pglist_data *pgdat, enum node_stat_item item) |
330 | { | |
331 | struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats; | |
332 | s8 __percpu *p = pcp->vm_node_stat_diff + item; | |
333 | s8 v, t; | |
334 | ||
335 | v = __this_cpu_inc_return(*p); | |
336 | t = __this_cpu_read(pcp->stat_threshold); | |
337 | if (unlikely(v > t)) { | |
338 | s8 overstep = t >> 1; | |
339 | ||
340 | node_page_state_add(v + overstep, pgdat, item); | |
341 | __this_cpu_write(*p, -overstep); | |
342 | } | |
343 | } | |
344 | ||
ca889e6c CL |
345 | void __inc_zone_page_state(struct page *page, enum zone_stat_item item) |
346 | { | |
347 | __inc_zone_state(page_zone(page), item); | |
348 | } | |
2244b95a CL |
349 | EXPORT_SYMBOL(__inc_zone_page_state); |
350 | ||
75ef7184 MG |
351 | void __inc_node_page_state(struct page *page, enum node_stat_item item) |
352 | { | |
353 | __inc_node_state(page_pgdat(page), item); | |
354 | } | |
355 | EXPORT_SYMBOL(__inc_node_page_state); | |
356 | ||
c8785385 | 357 | void __dec_zone_state(struct zone *zone, enum zone_stat_item item) |
2244b95a | 358 | { |
12938a92 CL |
359 | struct per_cpu_pageset __percpu *pcp = zone->pageset; |
360 | s8 __percpu *p = pcp->vm_stat_diff + item; | |
361 | s8 v, t; | |
2244b95a | 362 | |
908ee0f1 | 363 | v = __this_cpu_dec_return(*p); |
12938a92 CL |
364 | t = __this_cpu_read(pcp->stat_threshold); |
365 | if (unlikely(v < - t)) { | |
366 | s8 overstep = t >> 1; | |
2244b95a | 367 | |
12938a92 CL |
368 | zone_page_state_add(v - overstep, zone, item); |
369 | __this_cpu_write(*p, overstep); | |
2244b95a CL |
370 | } |
371 | } | |
c8785385 | 372 | |
75ef7184 MG |
373 | void __dec_node_state(struct pglist_data *pgdat, enum node_stat_item item) |
374 | { | |
375 | struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats; | |
376 | s8 __percpu *p = pcp->vm_node_stat_diff + item; | |
377 | s8 v, t; | |
378 | ||
379 | v = __this_cpu_dec_return(*p); | |
380 | t = __this_cpu_read(pcp->stat_threshold); | |
381 | if (unlikely(v < - t)) { | |
382 | s8 overstep = t >> 1; | |
383 | ||
384 | node_page_state_add(v - overstep, pgdat, item); | |
385 | __this_cpu_write(*p, overstep); | |
386 | } | |
387 | } | |
388 | ||
c8785385 CL |
389 | void __dec_zone_page_state(struct page *page, enum zone_stat_item item) |
390 | { | |
391 | __dec_zone_state(page_zone(page), item); | |
392 | } | |
2244b95a CL |
393 | EXPORT_SYMBOL(__dec_zone_page_state); |
394 | ||
75ef7184 MG |
395 | void __dec_node_page_state(struct page *page, enum node_stat_item item) |
396 | { | |
397 | __dec_node_state(page_pgdat(page), item); | |
398 | } | |
399 | EXPORT_SYMBOL(__dec_node_page_state); | |
400 | ||
4156153c | 401 | #ifdef CONFIG_HAVE_CMPXCHG_LOCAL |
7c839120 CL |
402 | /* |
403 | * If we have cmpxchg_local support then we do not need to incur the overhead | |
404 | * that comes with local_irq_save/restore if we use this_cpu_cmpxchg. | |
405 | * | |
406 | * mod_state() modifies the zone counter state through atomic per cpu | |
407 | * operations. | |
408 | * | |
409 | * Overstep mode specifies how overstep should handled: | |
410 | * 0 No overstepping | |
411 | * 1 Overstepping half of threshold | |
412 | * -1 Overstepping minus half of threshold | |
413 | */ | |
75ef7184 MG |
414 | static inline void mod_zone_state(struct zone *zone, |
415 | enum zone_stat_item item, long delta, int overstep_mode) | |
7c839120 CL |
416 | { |
417 | struct per_cpu_pageset __percpu *pcp = zone->pageset; | |
418 | s8 __percpu *p = pcp->vm_stat_diff + item; | |
419 | long o, n, t, z; | |
420 | ||
421 | do { | |
422 | z = 0; /* overflow to zone counters */ | |
423 | ||
424 | /* | |
425 | * The fetching of the stat_threshold is racy. We may apply | |
426 | * a counter threshold to the wrong the cpu if we get | |
d3bc2367 CL |
427 | * rescheduled while executing here. However, the next |
428 | * counter update will apply the threshold again and | |
429 | * therefore bring the counter under the threshold again. | |
430 | * | |
431 | * Most of the time the thresholds are the same anyways | |
432 | * for all cpus in a zone. | |
7c839120 CL |
433 | */ |
434 | t = this_cpu_read(pcp->stat_threshold); | |
435 | ||
436 | o = this_cpu_read(*p); | |
437 | n = delta + o; | |
438 | ||
439 | if (n > t || n < -t) { | |
440 | int os = overstep_mode * (t >> 1) ; | |
441 | ||
442 | /* Overflow must be added to zone counters */ | |
443 | z = n + os; | |
444 | n = -os; | |
445 | } | |
446 | } while (this_cpu_cmpxchg(*p, o, n) != o); | |
447 | ||
448 | if (z) | |
449 | zone_page_state_add(z, zone, item); | |
450 | } | |
451 | ||
452 | void mod_zone_page_state(struct zone *zone, enum zone_stat_item item, | |
6cdb18ad | 453 | long delta) |
7c839120 | 454 | { |
75ef7184 | 455 | mod_zone_state(zone, item, delta, 0); |
7c839120 CL |
456 | } |
457 | EXPORT_SYMBOL(mod_zone_page_state); | |
458 | ||
7c839120 CL |
459 | void inc_zone_page_state(struct page *page, enum zone_stat_item item) |
460 | { | |
75ef7184 | 461 | mod_zone_state(page_zone(page), item, 1, 1); |
7c839120 CL |
462 | } |
463 | EXPORT_SYMBOL(inc_zone_page_state); | |
464 | ||
465 | void dec_zone_page_state(struct page *page, enum zone_stat_item item) | |
466 | { | |
75ef7184 | 467 | mod_zone_state(page_zone(page), item, -1, -1); |
7c839120 CL |
468 | } |
469 | EXPORT_SYMBOL(dec_zone_page_state); | |
75ef7184 MG |
470 | |
471 | static inline void mod_node_state(struct pglist_data *pgdat, | |
472 | enum node_stat_item item, int delta, int overstep_mode) | |
473 | { | |
474 | struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats; | |
475 | s8 __percpu *p = pcp->vm_node_stat_diff + item; | |
476 | long o, n, t, z; | |
477 | ||
478 | do { | |
479 | z = 0; /* overflow to node counters */ | |
480 | ||
481 | /* | |
482 | * The fetching of the stat_threshold is racy. We may apply | |
483 | * a counter threshold to the wrong the cpu if we get | |
484 | * rescheduled while executing here. However, the next | |
485 | * counter update will apply the threshold again and | |
486 | * therefore bring the counter under the threshold again. | |
487 | * | |
488 | * Most of the time the thresholds are the same anyways | |
489 | * for all cpus in a node. | |
490 | */ | |
491 | t = this_cpu_read(pcp->stat_threshold); | |
492 | ||
493 | o = this_cpu_read(*p); | |
494 | n = delta + o; | |
495 | ||
496 | if (n > t || n < -t) { | |
497 | int os = overstep_mode * (t >> 1) ; | |
498 | ||
499 | /* Overflow must be added to node counters */ | |
500 | z = n + os; | |
501 | n = -os; | |
502 | } | |
503 | } while (this_cpu_cmpxchg(*p, o, n) != o); | |
504 | ||
505 | if (z) | |
506 | node_page_state_add(z, pgdat, item); | |
507 | } | |
508 | ||
509 | void mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item, | |
510 | long delta) | |
511 | { | |
512 | mod_node_state(pgdat, item, delta, 0); | |
513 | } | |
514 | EXPORT_SYMBOL(mod_node_page_state); | |
515 | ||
516 | void inc_node_state(struct pglist_data *pgdat, enum node_stat_item item) | |
517 | { | |
518 | mod_node_state(pgdat, item, 1, 1); | |
519 | } | |
520 | ||
521 | void inc_node_page_state(struct page *page, enum node_stat_item item) | |
522 | { | |
523 | mod_node_state(page_pgdat(page), item, 1, 1); | |
524 | } | |
525 | EXPORT_SYMBOL(inc_node_page_state); | |
526 | ||
527 | void dec_node_page_state(struct page *page, enum node_stat_item item) | |
528 | { | |
529 | mod_node_state(page_pgdat(page), item, -1, -1); | |
530 | } | |
531 | EXPORT_SYMBOL(dec_node_page_state); | |
7c839120 CL |
532 | #else |
533 | /* | |
534 | * Use interrupt disable to serialize counter updates | |
535 | */ | |
536 | void mod_zone_page_state(struct zone *zone, enum zone_stat_item item, | |
6cdb18ad | 537 | long delta) |
7c839120 CL |
538 | { |
539 | unsigned long flags; | |
540 | ||
541 | local_irq_save(flags); | |
542 | __mod_zone_page_state(zone, item, delta); | |
543 | local_irq_restore(flags); | |
544 | } | |
545 | EXPORT_SYMBOL(mod_zone_page_state); | |
546 | ||
2244b95a CL |
547 | void inc_zone_page_state(struct page *page, enum zone_stat_item item) |
548 | { | |
549 | unsigned long flags; | |
550 | struct zone *zone; | |
2244b95a CL |
551 | |
552 | zone = page_zone(page); | |
553 | local_irq_save(flags); | |
ca889e6c | 554 | __inc_zone_state(zone, item); |
2244b95a CL |
555 | local_irq_restore(flags); |
556 | } | |
557 | EXPORT_SYMBOL(inc_zone_page_state); | |
558 | ||
559 | void dec_zone_page_state(struct page *page, enum zone_stat_item item) | |
560 | { | |
561 | unsigned long flags; | |
2244b95a | 562 | |
2244b95a | 563 | local_irq_save(flags); |
a302eb4e | 564 | __dec_zone_page_state(page, item); |
2244b95a CL |
565 | local_irq_restore(flags); |
566 | } | |
567 | EXPORT_SYMBOL(dec_zone_page_state); | |
568 | ||
75ef7184 MG |
569 | void inc_node_state(struct pglist_data *pgdat, enum node_stat_item item) |
570 | { | |
571 | unsigned long flags; | |
572 | ||
573 | local_irq_save(flags); | |
574 | __inc_node_state(pgdat, item); | |
575 | local_irq_restore(flags); | |
576 | } | |
577 | EXPORT_SYMBOL(inc_node_state); | |
578 | ||
579 | void mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item, | |
580 | long delta) | |
581 | { | |
582 | unsigned long flags; | |
583 | ||
584 | local_irq_save(flags); | |
585 | __mod_node_page_state(pgdat, item, delta); | |
586 | local_irq_restore(flags); | |
587 | } | |
588 | EXPORT_SYMBOL(mod_node_page_state); | |
589 | ||
590 | void inc_node_page_state(struct page *page, enum node_stat_item item) | |
591 | { | |
592 | unsigned long flags; | |
593 | struct pglist_data *pgdat; | |
594 | ||
595 | pgdat = page_pgdat(page); | |
596 | local_irq_save(flags); | |
597 | __inc_node_state(pgdat, item); | |
598 | local_irq_restore(flags); | |
599 | } | |
600 | EXPORT_SYMBOL(inc_node_page_state); | |
601 | ||
602 | void dec_node_page_state(struct page *page, enum node_stat_item item) | |
603 | { | |
604 | unsigned long flags; | |
605 | ||
606 | local_irq_save(flags); | |
607 | __dec_node_page_state(page, item); | |
608 | local_irq_restore(flags); | |
609 | } | |
610 | EXPORT_SYMBOL(dec_node_page_state); | |
611 | #endif | |
7cc36bbd CL |
612 | |
613 | /* | |
614 | * Fold a differential into the global counters. | |
615 | * Returns the number of counters updated. | |
616 | */ | |
3a321d2a KW |
617 | #ifdef CONFIG_NUMA |
618 | static int fold_diff(int *zone_diff, int *numa_diff, int *node_diff) | |
619 | { | |
620 | int i; | |
621 | int changes = 0; | |
622 | ||
623 | for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) | |
624 | if (zone_diff[i]) { | |
625 | atomic_long_add(zone_diff[i], &vm_zone_stat[i]); | |
626 | changes++; | |
627 | } | |
628 | ||
629 | for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++) | |
630 | if (numa_diff[i]) { | |
631 | atomic_long_add(numa_diff[i], &vm_numa_stat[i]); | |
632 | changes++; | |
633 | } | |
634 | ||
635 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) | |
636 | if (node_diff[i]) { | |
637 | atomic_long_add(node_diff[i], &vm_node_stat[i]); | |
638 | changes++; | |
639 | } | |
640 | return changes; | |
641 | } | |
642 | #else | |
75ef7184 | 643 | static int fold_diff(int *zone_diff, int *node_diff) |
4edb0748 CL |
644 | { |
645 | int i; | |
7cc36bbd | 646 | int changes = 0; |
4edb0748 CL |
647 | |
648 | for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) | |
75ef7184 MG |
649 | if (zone_diff[i]) { |
650 | atomic_long_add(zone_diff[i], &vm_zone_stat[i]); | |
651 | changes++; | |
652 | } | |
653 | ||
654 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) | |
655 | if (node_diff[i]) { | |
656 | atomic_long_add(node_diff[i], &vm_node_stat[i]); | |
7cc36bbd CL |
657 | changes++; |
658 | } | |
659 | return changes; | |
4edb0748 | 660 | } |
3a321d2a | 661 | #endif /* CONFIG_NUMA */ |
4edb0748 | 662 | |
2244b95a | 663 | /* |
2bb921e5 | 664 | * Update the zone counters for the current cpu. |
a7f75e25 | 665 | * |
4037d452 CL |
666 | * Note that refresh_cpu_vm_stats strives to only access |
667 | * node local memory. The per cpu pagesets on remote zones are placed | |
668 | * in the memory local to the processor using that pageset. So the | |
669 | * loop over all zones will access a series of cachelines local to | |
670 | * the processor. | |
671 | * | |
672 | * The call to zone_page_state_add updates the cachelines with the | |
673 | * statistics in the remote zone struct as well as the global cachelines | |
674 | * with the global counters. These could cause remote node cache line | |
675 | * bouncing and will have to be only done when necessary. | |
7cc36bbd CL |
676 | * |
677 | * The function returns the number of global counters updated. | |
2244b95a | 678 | */ |
0eb77e98 | 679 | static int refresh_cpu_vm_stats(bool do_pagesets) |
2244b95a | 680 | { |
75ef7184 | 681 | struct pglist_data *pgdat; |
2244b95a CL |
682 | struct zone *zone; |
683 | int i; | |
75ef7184 | 684 | int global_zone_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, }; |
3a321d2a KW |
685 | #ifdef CONFIG_NUMA |
686 | int global_numa_diff[NR_VM_NUMA_STAT_ITEMS] = { 0, }; | |
687 | #endif | |
75ef7184 | 688 | int global_node_diff[NR_VM_NODE_STAT_ITEMS] = { 0, }; |
7cc36bbd | 689 | int changes = 0; |
2244b95a | 690 | |
ee99c71c | 691 | for_each_populated_zone(zone) { |
fbc2edb0 | 692 | struct per_cpu_pageset __percpu *p = zone->pageset; |
2244b95a | 693 | |
fbc2edb0 CL |
694 | for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) { |
695 | int v; | |
2244b95a | 696 | |
fbc2edb0 CL |
697 | v = this_cpu_xchg(p->vm_stat_diff[i], 0); |
698 | if (v) { | |
a7f75e25 | 699 | |
a7f75e25 | 700 | atomic_long_add(v, &zone->vm_stat[i]); |
75ef7184 | 701 | global_zone_diff[i] += v; |
4037d452 CL |
702 | #ifdef CONFIG_NUMA |
703 | /* 3 seconds idle till flush */ | |
fbc2edb0 | 704 | __this_cpu_write(p->expire, 3); |
4037d452 | 705 | #endif |
2244b95a | 706 | } |
fbc2edb0 | 707 | } |
4037d452 | 708 | #ifdef CONFIG_NUMA |
3a321d2a KW |
709 | for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++) { |
710 | int v; | |
711 | ||
712 | v = this_cpu_xchg(p->vm_numa_stat_diff[i], 0); | |
713 | if (v) { | |
714 | ||
715 | atomic_long_add(v, &zone->vm_numa_stat[i]); | |
716 | global_numa_diff[i] += v; | |
717 | __this_cpu_write(p->expire, 3); | |
718 | } | |
719 | } | |
720 | ||
0eb77e98 CL |
721 | if (do_pagesets) { |
722 | cond_resched(); | |
723 | /* | |
724 | * Deal with draining the remote pageset of this | |
725 | * processor | |
726 | * | |
727 | * Check if there are pages remaining in this pageset | |
728 | * if not then there is nothing to expire. | |
729 | */ | |
730 | if (!__this_cpu_read(p->expire) || | |
fbc2edb0 | 731 | !__this_cpu_read(p->pcp.count)) |
0eb77e98 | 732 | continue; |
4037d452 | 733 | |
0eb77e98 CL |
734 | /* |
735 | * We never drain zones local to this processor. | |
736 | */ | |
737 | if (zone_to_nid(zone) == numa_node_id()) { | |
738 | __this_cpu_write(p->expire, 0); | |
739 | continue; | |
740 | } | |
4037d452 | 741 | |
0eb77e98 CL |
742 | if (__this_cpu_dec_return(p->expire)) |
743 | continue; | |
4037d452 | 744 | |
0eb77e98 CL |
745 | if (__this_cpu_read(p->pcp.count)) { |
746 | drain_zone_pages(zone, this_cpu_ptr(&p->pcp)); | |
747 | changes++; | |
748 | } | |
7cc36bbd | 749 | } |
4037d452 | 750 | #endif |
2244b95a | 751 | } |
75ef7184 MG |
752 | |
753 | for_each_online_pgdat(pgdat) { | |
754 | struct per_cpu_nodestat __percpu *p = pgdat->per_cpu_nodestats; | |
755 | ||
756 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) { | |
757 | int v; | |
758 | ||
759 | v = this_cpu_xchg(p->vm_node_stat_diff[i], 0); | |
760 | if (v) { | |
761 | atomic_long_add(v, &pgdat->vm_stat[i]); | |
762 | global_node_diff[i] += v; | |
763 | } | |
764 | } | |
765 | } | |
766 | ||
3a321d2a KW |
767 | #ifdef CONFIG_NUMA |
768 | changes += fold_diff(global_zone_diff, global_numa_diff, | |
769 | global_node_diff); | |
770 | #else | |
75ef7184 | 771 | changes += fold_diff(global_zone_diff, global_node_diff); |
3a321d2a | 772 | #endif |
7cc36bbd | 773 | return changes; |
2244b95a CL |
774 | } |
775 | ||
2bb921e5 CL |
776 | /* |
777 | * Fold the data for an offline cpu into the global array. | |
778 | * There cannot be any access by the offline cpu and therefore | |
779 | * synchronization is simplified. | |
780 | */ | |
781 | void cpu_vm_stats_fold(int cpu) | |
782 | { | |
75ef7184 | 783 | struct pglist_data *pgdat; |
2bb921e5 CL |
784 | struct zone *zone; |
785 | int i; | |
75ef7184 | 786 | int global_zone_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, }; |
3a321d2a KW |
787 | #ifdef CONFIG_NUMA |
788 | int global_numa_diff[NR_VM_NUMA_STAT_ITEMS] = { 0, }; | |
789 | #endif | |
75ef7184 | 790 | int global_node_diff[NR_VM_NODE_STAT_ITEMS] = { 0, }; |
2bb921e5 CL |
791 | |
792 | for_each_populated_zone(zone) { | |
793 | struct per_cpu_pageset *p; | |
794 | ||
795 | p = per_cpu_ptr(zone->pageset, cpu); | |
796 | ||
797 | for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) | |
798 | if (p->vm_stat_diff[i]) { | |
799 | int v; | |
800 | ||
801 | v = p->vm_stat_diff[i]; | |
802 | p->vm_stat_diff[i] = 0; | |
803 | atomic_long_add(v, &zone->vm_stat[i]); | |
75ef7184 | 804 | global_zone_diff[i] += v; |
2bb921e5 | 805 | } |
3a321d2a KW |
806 | |
807 | #ifdef CONFIG_NUMA | |
808 | for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++) | |
809 | if (p->vm_numa_stat_diff[i]) { | |
810 | int v; | |
811 | ||
812 | v = p->vm_numa_stat_diff[i]; | |
813 | p->vm_numa_stat_diff[i] = 0; | |
814 | atomic_long_add(v, &zone->vm_numa_stat[i]); | |
815 | global_numa_diff[i] += v; | |
816 | } | |
817 | #endif | |
2bb921e5 CL |
818 | } |
819 | ||
75ef7184 MG |
820 | for_each_online_pgdat(pgdat) { |
821 | struct per_cpu_nodestat *p; | |
822 | ||
823 | p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu); | |
824 | ||
825 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) | |
826 | if (p->vm_node_stat_diff[i]) { | |
827 | int v; | |
828 | ||
829 | v = p->vm_node_stat_diff[i]; | |
830 | p->vm_node_stat_diff[i] = 0; | |
831 | atomic_long_add(v, &pgdat->vm_stat[i]); | |
832 | global_node_diff[i] += v; | |
833 | } | |
834 | } | |
835 | ||
3a321d2a KW |
836 | #ifdef CONFIG_NUMA |
837 | fold_diff(global_zone_diff, global_numa_diff, global_node_diff); | |
838 | #else | |
75ef7184 | 839 | fold_diff(global_zone_diff, global_node_diff); |
3a321d2a | 840 | #endif |
2bb921e5 CL |
841 | } |
842 | ||
40f4b1ea CS |
843 | /* |
844 | * this is only called if !populated_zone(zone), which implies no other users of | |
845 | * pset->vm_stat_diff[] exsist. | |
846 | */ | |
5a883813 MK |
847 | void drain_zonestat(struct zone *zone, struct per_cpu_pageset *pset) |
848 | { | |
849 | int i; | |
850 | ||
851 | for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) | |
852 | if (pset->vm_stat_diff[i]) { | |
853 | int v = pset->vm_stat_diff[i]; | |
854 | pset->vm_stat_diff[i] = 0; | |
855 | atomic_long_add(v, &zone->vm_stat[i]); | |
75ef7184 | 856 | atomic_long_add(v, &vm_zone_stat[i]); |
5a883813 | 857 | } |
3a321d2a KW |
858 | |
859 | #ifdef CONFIG_NUMA | |
860 | for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++) | |
861 | if (pset->vm_numa_stat_diff[i]) { | |
862 | int v = pset->vm_numa_stat_diff[i]; | |
863 | ||
864 | pset->vm_numa_stat_diff[i] = 0; | |
865 | atomic_long_add(v, &zone->vm_numa_stat[i]); | |
866 | atomic_long_add(v, &vm_numa_stat[i]); | |
867 | } | |
868 | #endif | |
5a883813 | 869 | } |
2244b95a CL |
870 | #endif |
871 | ||
ca889e6c | 872 | #ifdef CONFIG_NUMA |
3a321d2a KW |
873 | void __inc_numa_state(struct zone *zone, |
874 | enum numa_stat_item item) | |
875 | { | |
876 | struct per_cpu_pageset __percpu *pcp = zone->pageset; | |
877 | s8 __percpu *p = pcp->vm_numa_stat_diff + item; | |
878 | s8 v, t; | |
879 | ||
880 | v = __this_cpu_inc_return(*p); | |
881 | t = __this_cpu_read(pcp->numa_stat_threshold); | |
882 | if (unlikely(v > t)) { | |
883 | s8 overstep = t >> 1; | |
884 | ||
885 | zone_numa_state_add(v + overstep, zone, item); | |
886 | __this_cpu_write(*p, -overstep); | |
887 | } | |
888 | } | |
889 | ||
c2d42c16 | 890 | /* |
75ef7184 MG |
891 | * Determine the per node value of a stat item. This function |
892 | * is called frequently in a NUMA machine, so try to be as | |
893 | * frugal as possible. | |
c2d42c16 | 894 | */ |
75ef7184 MG |
895 | unsigned long sum_zone_node_page_state(int node, |
896 | enum zone_stat_item item) | |
c2d42c16 AM |
897 | { |
898 | struct zone *zones = NODE_DATA(node)->node_zones; | |
e87d59f7 JK |
899 | int i; |
900 | unsigned long count = 0; | |
c2d42c16 | 901 | |
e87d59f7 JK |
902 | for (i = 0; i < MAX_NR_ZONES; i++) |
903 | count += zone_page_state(zones + i, item); | |
904 | ||
905 | return count; | |
c2d42c16 AM |
906 | } |
907 | ||
3a321d2a KW |
908 | unsigned long sum_zone_numa_state(int node, |
909 | enum numa_stat_item item) | |
910 | { | |
911 | struct zone *zones = NODE_DATA(node)->node_zones; | |
912 | int i; | |
913 | unsigned long count = 0; | |
914 | ||
915 | for (i = 0; i < MAX_NR_ZONES; i++) | |
916 | count += zone_numa_state(zones + i, item); | |
917 | ||
918 | return count; | |
919 | } | |
920 | ||
75ef7184 MG |
921 | /* |
922 | * Determine the per node value of a stat item. | |
923 | */ | |
924 | unsigned long node_page_state(struct pglist_data *pgdat, | |
925 | enum node_stat_item item) | |
926 | { | |
927 | long x = atomic_long_read(&pgdat->vm_stat[item]); | |
928 | #ifdef CONFIG_SMP | |
929 | if (x < 0) | |
930 | x = 0; | |
931 | #endif | |
932 | return x; | |
933 | } | |
ca889e6c CL |
934 | #endif |
935 | ||
d7a5752c | 936 | #ifdef CONFIG_COMPACTION |
36deb0be | 937 | |
d7a5752c MG |
938 | struct contig_page_info { |
939 | unsigned long free_pages; | |
940 | unsigned long free_blocks_total; | |
941 | unsigned long free_blocks_suitable; | |
942 | }; | |
943 | ||
944 | /* | |
945 | * Calculate the number of free pages in a zone, how many contiguous | |
946 | * pages are free and how many are large enough to satisfy an allocation of | |
947 | * the target size. Note that this function makes no attempt to estimate | |
948 | * how many suitable free blocks there *might* be if MOVABLE pages were | |
949 | * migrated. Calculating that is possible, but expensive and can be | |
950 | * figured out from userspace | |
951 | */ | |
952 | static void fill_contig_page_info(struct zone *zone, | |
953 | unsigned int suitable_order, | |
954 | struct contig_page_info *info) | |
955 | { | |
956 | unsigned int order; | |
957 | ||
958 | info->free_pages = 0; | |
959 | info->free_blocks_total = 0; | |
960 | info->free_blocks_suitable = 0; | |
961 | ||
962 | for (order = 0; order < MAX_ORDER; order++) { | |
963 | unsigned long blocks; | |
964 | ||
965 | /* Count number of free blocks */ | |
966 | blocks = zone->free_area[order].nr_free; | |
967 | info->free_blocks_total += blocks; | |
968 | ||
969 | /* Count free base pages */ | |
970 | info->free_pages += blocks << order; | |
971 | ||
972 | /* Count the suitable free blocks */ | |
973 | if (order >= suitable_order) | |
974 | info->free_blocks_suitable += blocks << | |
975 | (order - suitable_order); | |
976 | } | |
977 | } | |
f1a5ab12 MG |
978 | |
979 | /* | |
980 | * A fragmentation index only makes sense if an allocation of a requested | |
981 | * size would fail. If that is true, the fragmentation index indicates | |
982 | * whether external fragmentation or a lack of memory was the problem. | |
983 | * The value can be used to determine if page reclaim or compaction | |
984 | * should be used | |
985 | */ | |
56de7263 | 986 | static int __fragmentation_index(unsigned int order, struct contig_page_info *info) |
f1a5ab12 MG |
987 | { |
988 | unsigned long requested = 1UL << order; | |
989 | ||
88d6ac40 WY |
990 | if (WARN_ON_ONCE(order >= MAX_ORDER)) |
991 | return 0; | |
992 | ||
f1a5ab12 MG |
993 | if (!info->free_blocks_total) |
994 | return 0; | |
995 | ||
996 | /* Fragmentation index only makes sense when a request would fail */ | |
997 | if (info->free_blocks_suitable) | |
998 | return -1000; | |
999 | ||
1000 | /* | |
1001 | * Index is between 0 and 1 so return within 3 decimal places | |
1002 | * | |
1003 | * 0 => allocation would fail due to lack of memory | |
1004 | * 1 => allocation would fail due to fragmentation | |
1005 | */ | |
1006 | return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total); | |
1007 | } | |
56de7263 MG |
1008 | |
1009 | /* Same as __fragmentation index but allocs contig_page_info on stack */ | |
1010 | int fragmentation_index(struct zone *zone, unsigned int order) | |
1011 | { | |
1012 | struct contig_page_info info; | |
1013 | ||
1014 | fill_contig_page_info(zone, order, &info); | |
1015 | return __fragmentation_index(order, &info); | |
1016 | } | |
d7a5752c MG |
1017 | #endif |
1018 | ||
0d6617c7 | 1019 | #if defined(CONFIG_PROC_FS) || defined(CONFIG_SYSFS) || defined(CONFIG_NUMA) |
fa25c503 KM |
1020 | #ifdef CONFIG_ZONE_DMA |
1021 | #define TEXT_FOR_DMA(xx) xx "_dma", | |
1022 | #else | |
1023 | #define TEXT_FOR_DMA(xx) | |
1024 | #endif | |
1025 | ||
1026 | #ifdef CONFIG_ZONE_DMA32 | |
1027 | #define TEXT_FOR_DMA32(xx) xx "_dma32", | |
1028 | #else | |
1029 | #define TEXT_FOR_DMA32(xx) | |
1030 | #endif | |
1031 | ||
1032 | #ifdef CONFIG_HIGHMEM | |
1033 | #define TEXT_FOR_HIGHMEM(xx) xx "_high", | |
1034 | #else | |
1035 | #define TEXT_FOR_HIGHMEM(xx) | |
1036 | #endif | |
1037 | ||
1038 | #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \ | |
1039 | TEXT_FOR_HIGHMEM(xx) xx "_movable", | |
1040 | ||
1041 | const char * const vmstat_text[] = { | |
09316c09 | 1042 | /* enum zone_stat_item countes */ |
fa25c503 | 1043 | "nr_free_pages", |
71c799f4 MK |
1044 | "nr_zone_inactive_anon", |
1045 | "nr_zone_active_anon", | |
1046 | "nr_zone_inactive_file", | |
1047 | "nr_zone_active_file", | |
1048 | "nr_zone_unevictable", | |
5a1c84b4 | 1049 | "nr_zone_write_pending", |
fa25c503 | 1050 | "nr_mlock", |
fa25c503 KM |
1051 | "nr_page_table_pages", |
1052 | "nr_kernel_stack", | |
fa25c503 | 1053 | "nr_bounce", |
91537fee MK |
1054 | #if IS_ENABLED(CONFIG_ZSMALLOC) |
1055 | "nr_zspages", | |
1056 | #endif | |
3a321d2a KW |
1057 | "nr_free_cma", |
1058 | ||
1059 | /* enum numa_stat_item counters */ | |
fa25c503 KM |
1060 | #ifdef CONFIG_NUMA |
1061 | "numa_hit", | |
1062 | "numa_miss", | |
1063 | "numa_foreign", | |
1064 | "numa_interleave", | |
1065 | "numa_local", | |
1066 | "numa_other", | |
1067 | #endif | |
09316c09 | 1068 | |
599d0c95 MG |
1069 | /* Node-based counters */ |
1070 | "nr_inactive_anon", | |
1071 | "nr_active_anon", | |
1072 | "nr_inactive_file", | |
1073 | "nr_active_file", | |
1074 | "nr_unevictable", | |
385386cf JW |
1075 | "nr_slab_reclaimable", |
1076 | "nr_slab_unreclaimable", | |
599d0c95 MG |
1077 | "nr_isolated_anon", |
1078 | "nr_isolated_file", | |
1e6b1085 MG |
1079 | "workingset_refault", |
1080 | "workingset_activate", | |
1081 | "workingset_nodereclaim", | |
50658e2e MG |
1082 | "nr_anon_pages", |
1083 | "nr_mapped", | |
11fb9989 MG |
1084 | "nr_file_pages", |
1085 | "nr_dirty", | |
1086 | "nr_writeback", | |
1087 | "nr_writeback_temp", | |
1088 | "nr_shmem", | |
1089 | "nr_shmem_hugepages", | |
1090 | "nr_shmem_pmdmapped", | |
1091 | "nr_anon_transparent_hugepages", | |
1092 | "nr_unstable", | |
c4a25635 MG |
1093 | "nr_vmscan_write", |
1094 | "nr_vmscan_immediate_reclaim", | |
1095 | "nr_dirtied", | |
1096 | "nr_written", | |
599d0c95 | 1097 | |
09316c09 | 1098 | /* enum writeback_stat_item counters */ |
fa25c503 KM |
1099 | "nr_dirty_threshold", |
1100 | "nr_dirty_background_threshold", | |
1101 | ||
1102 | #ifdef CONFIG_VM_EVENT_COUNTERS | |
09316c09 | 1103 | /* enum vm_event_item counters */ |
fa25c503 KM |
1104 | "pgpgin", |
1105 | "pgpgout", | |
1106 | "pswpin", | |
1107 | "pswpout", | |
1108 | ||
1109 | TEXTS_FOR_ZONES("pgalloc") | |
7cc30fcf MG |
1110 | TEXTS_FOR_ZONES("allocstall") |
1111 | TEXTS_FOR_ZONES("pgskip") | |
fa25c503 KM |
1112 | |
1113 | "pgfree", | |
1114 | "pgactivate", | |
1115 | "pgdeactivate", | |
f7ad2a6c | 1116 | "pglazyfree", |
fa25c503 KM |
1117 | |
1118 | "pgfault", | |
1119 | "pgmajfault", | |
854e9ed0 | 1120 | "pglazyfreed", |
fa25c503 | 1121 | |
599d0c95 MG |
1122 | "pgrefill", |
1123 | "pgsteal_kswapd", | |
1124 | "pgsteal_direct", | |
1125 | "pgscan_kswapd", | |
1126 | "pgscan_direct", | |
68243e76 | 1127 | "pgscan_direct_throttle", |
fa25c503 KM |
1128 | |
1129 | #ifdef CONFIG_NUMA | |
1130 | "zone_reclaim_failed", | |
1131 | #endif | |
1132 | "pginodesteal", | |
1133 | "slabs_scanned", | |
fa25c503 KM |
1134 | "kswapd_inodesteal", |
1135 | "kswapd_low_wmark_hit_quickly", | |
1136 | "kswapd_high_wmark_hit_quickly", | |
fa25c503 | 1137 | "pageoutrun", |
fa25c503 KM |
1138 | |
1139 | "pgrotated", | |
1140 | ||
5509a5d2 DH |
1141 | "drop_pagecache", |
1142 | "drop_slab", | |
8e675f7a | 1143 | "oom_kill", |
5509a5d2 | 1144 | |
03c5a6e1 MG |
1145 | #ifdef CONFIG_NUMA_BALANCING |
1146 | "numa_pte_updates", | |
72403b4a | 1147 | "numa_huge_pte_updates", |
03c5a6e1 MG |
1148 | "numa_hint_faults", |
1149 | "numa_hint_faults_local", | |
1150 | "numa_pages_migrated", | |
1151 | #endif | |
5647bc29 MG |
1152 | #ifdef CONFIG_MIGRATION |
1153 | "pgmigrate_success", | |
1154 | "pgmigrate_fail", | |
1155 | #endif | |
fa25c503 | 1156 | #ifdef CONFIG_COMPACTION |
397487db MG |
1157 | "compact_migrate_scanned", |
1158 | "compact_free_scanned", | |
1159 | "compact_isolated", | |
fa25c503 KM |
1160 | "compact_stall", |
1161 | "compact_fail", | |
1162 | "compact_success", | |
698b1b30 | 1163 | "compact_daemon_wake", |
7f354a54 DR |
1164 | "compact_daemon_migrate_scanned", |
1165 | "compact_daemon_free_scanned", | |
fa25c503 KM |
1166 | #endif |
1167 | ||
1168 | #ifdef CONFIG_HUGETLB_PAGE | |
1169 | "htlb_buddy_alloc_success", | |
1170 | "htlb_buddy_alloc_fail", | |
1171 | #endif | |
1172 | "unevictable_pgs_culled", | |
1173 | "unevictable_pgs_scanned", | |
1174 | "unevictable_pgs_rescued", | |
1175 | "unevictable_pgs_mlocked", | |
1176 | "unevictable_pgs_munlocked", | |
1177 | "unevictable_pgs_cleared", | |
1178 | "unevictable_pgs_stranded", | |
fa25c503 KM |
1179 | |
1180 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
1181 | "thp_fault_alloc", | |
1182 | "thp_fault_fallback", | |
1183 | "thp_collapse_alloc", | |
1184 | "thp_collapse_alloc_failed", | |
95ecedcd KS |
1185 | "thp_file_alloc", |
1186 | "thp_file_mapped", | |
122afea9 KS |
1187 | "thp_split_page", |
1188 | "thp_split_page_failed", | |
f9719a03 | 1189 | "thp_deferred_split_page", |
122afea9 | 1190 | "thp_split_pmd", |
ce9311cf YX |
1191 | #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD |
1192 | "thp_split_pud", | |
1193 | #endif | |
d8a8e1f0 KS |
1194 | "thp_zero_page_alloc", |
1195 | "thp_zero_page_alloc_failed", | |
225311a4 | 1196 | "thp_swpout", |
fe490cc0 | 1197 | "thp_swpout_fallback", |
fa25c503 | 1198 | #endif |
09316c09 KK |
1199 | #ifdef CONFIG_MEMORY_BALLOON |
1200 | "balloon_inflate", | |
1201 | "balloon_deflate", | |
1202 | #ifdef CONFIG_BALLOON_COMPACTION | |
1203 | "balloon_migrate", | |
1204 | #endif | |
1205 | #endif /* CONFIG_MEMORY_BALLOON */ | |
ec659934 | 1206 | #ifdef CONFIG_DEBUG_TLBFLUSH |
6df46865 | 1207 | #ifdef CONFIG_SMP |
9824cf97 DH |
1208 | "nr_tlb_remote_flush", |
1209 | "nr_tlb_remote_flush_received", | |
ec659934 | 1210 | #endif /* CONFIG_SMP */ |
9824cf97 DH |
1211 | "nr_tlb_local_flush_all", |
1212 | "nr_tlb_local_flush_one", | |
ec659934 | 1213 | #endif /* CONFIG_DEBUG_TLBFLUSH */ |
fa25c503 | 1214 | |
4f115147 DB |
1215 | #ifdef CONFIG_DEBUG_VM_VMACACHE |
1216 | "vmacache_find_calls", | |
1217 | "vmacache_find_hits", | |
f5f302e2 | 1218 | "vmacache_full_flushes", |
4f115147 | 1219 | #endif |
cbc65df2 HY |
1220 | #ifdef CONFIG_SWAP |
1221 | "swap_ra", | |
1222 | "swap_ra_hit", | |
1223 | #endif | |
fa25c503 KM |
1224 | #endif /* CONFIG_VM_EVENTS_COUNTERS */ |
1225 | }; | |
0d6617c7 | 1226 | #endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA */ |
fa25c503 | 1227 | |
3c486871 AM |
1228 | #if (defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)) || \ |
1229 | defined(CONFIG_PROC_FS) | |
1230 | static void *frag_start(struct seq_file *m, loff_t *pos) | |
1231 | { | |
1232 | pg_data_t *pgdat; | |
1233 | loff_t node = *pos; | |
1234 | ||
1235 | for (pgdat = first_online_pgdat(); | |
1236 | pgdat && node; | |
1237 | pgdat = next_online_pgdat(pgdat)) | |
1238 | --node; | |
1239 | ||
1240 | return pgdat; | |
1241 | } | |
1242 | ||
1243 | static void *frag_next(struct seq_file *m, void *arg, loff_t *pos) | |
1244 | { | |
1245 | pg_data_t *pgdat = (pg_data_t *)arg; | |
1246 | ||
1247 | (*pos)++; | |
1248 | return next_online_pgdat(pgdat); | |
1249 | } | |
1250 | ||
1251 | static void frag_stop(struct seq_file *m, void *arg) | |
1252 | { | |
1253 | } | |
1254 | ||
b2bd8598 DR |
1255 | /* |
1256 | * Walk zones in a node and print using a callback. | |
1257 | * If @assert_populated is true, only use callback for zones that are populated. | |
1258 | */ | |
3c486871 | 1259 | static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat, |
727c080f | 1260 | bool assert_populated, bool nolock, |
3c486871 AM |
1261 | void (*print)(struct seq_file *m, pg_data_t *, struct zone *)) |
1262 | { | |
1263 | struct zone *zone; | |
1264 | struct zone *node_zones = pgdat->node_zones; | |
1265 | unsigned long flags; | |
1266 | ||
1267 | for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) { | |
b2bd8598 | 1268 | if (assert_populated && !populated_zone(zone)) |
3c486871 AM |
1269 | continue; |
1270 | ||
727c080f VM |
1271 | if (!nolock) |
1272 | spin_lock_irqsave(&zone->lock, flags); | |
3c486871 | 1273 | print(m, pgdat, zone); |
727c080f VM |
1274 | if (!nolock) |
1275 | spin_unlock_irqrestore(&zone->lock, flags); | |
3c486871 AM |
1276 | } |
1277 | } | |
1278 | #endif | |
1279 | ||
d7a5752c | 1280 | #ifdef CONFIG_PROC_FS |
467c996c MG |
1281 | static void frag_show_print(struct seq_file *m, pg_data_t *pgdat, |
1282 | struct zone *zone) | |
1283 | { | |
1284 | int order; | |
1285 | ||
1286 | seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name); | |
1287 | for (order = 0; order < MAX_ORDER; ++order) | |
1288 | seq_printf(m, "%6lu ", zone->free_area[order].nr_free); | |
1289 | seq_putc(m, '\n'); | |
1290 | } | |
1291 | ||
1292 | /* | |
1293 | * This walks the free areas for each zone. | |
1294 | */ | |
1295 | static int frag_show(struct seq_file *m, void *arg) | |
1296 | { | |
1297 | pg_data_t *pgdat = (pg_data_t *)arg; | |
727c080f | 1298 | walk_zones_in_node(m, pgdat, true, false, frag_show_print); |
467c996c MG |
1299 | return 0; |
1300 | } | |
1301 | ||
1302 | static void pagetypeinfo_showfree_print(struct seq_file *m, | |
1303 | pg_data_t *pgdat, struct zone *zone) | |
1304 | { | |
1305 | int order, mtype; | |
1306 | ||
1307 | for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) { | |
1308 | seq_printf(m, "Node %4d, zone %8s, type %12s ", | |
1309 | pgdat->node_id, | |
1310 | zone->name, | |
1311 | migratetype_names[mtype]); | |
1312 | for (order = 0; order < MAX_ORDER; ++order) { | |
1313 | unsigned long freecount = 0; | |
1314 | struct free_area *area; | |
1315 | struct list_head *curr; | |
1316 | ||
1317 | area = &(zone->free_area[order]); | |
1318 | ||
1319 | list_for_each(curr, &area->free_list[mtype]) | |
1320 | freecount++; | |
1321 | seq_printf(m, "%6lu ", freecount); | |
1322 | } | |
f6ac2354 CL |
1323 | seq_putc(m, '\n'); |
1324 | } | |
467c996c MG |
1325 | } |
1326 | ||
1327 | /* Print out the free pages at each order for each migatetype */ | |
1328 | static int pagetypeinfo_showfree(struct seq_file *m, void *arg) | |
1329 | { | |
1330 | int order; | |
1331 | pg_data_t *pgdat = (pg_data_t *)arg; | |
1332 | ||
1333 | /* Print header */ | |
1334 | seq_printf(m, "%-43s ", "Free pages count per migrate type at order"); | |
1335 | for (order = 0; order < MAX_ORDER; ++order) | |
1336 | seq_printf(m, "%6d ", order); | |
1337 | seq_putc(m, '\n'); | |
1338 | ||
727c080f | 1339 | walk_zones_in_node(m, pgdat, true, false, pagetypeinfo_showfree_print); |
467c996c MG |
1340 | |
1341 | return 0; | |
1342 | } | |
1343 | ||
1344 | static void pagetypeinfo_showblockcount_print(struct seq_file *m, | |
1345 | pg_data_t *pgdat, struct zone *zone) | |
1346 | { | |
1347 | int mtype; | |
1348 | unsigned long pfn; | |
1349 | unsigned long start_pfn = zone->zone_start_pfn; | |
108bcc96 | 1350 | unsigned long end_pfn = zone_end_pfn(zone); |
467c996c MG |
1351 | unsigned long count[MIGRATE_TYPES] = { 0, }; |
1352 | ||
1353 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { | |
1354 | struct page *page; | |
1355 | ||
d336e94e MH |
1356 | page = pfn_to_online_page(pfn); |
1357 | if (!page) | |
467c996c MG |
1358 | continue; |
1359 | ||
eb33575c MG |
1360 | /* Watch for unexpected holes punched in the memmap */ |
1361 | if (!memmap_valid_within(pfn, page, zone)) | |
e80d6a24 | 1362 | continue; |
eb33575c | 1363 | |
a91c43c7 JK |
1364 | if (page_zone(page) != zone) |
1365 | continue; | |
1366 | ||
467c996c MG |
1367 | mtype = get_pageblock_migratetype(page); |
1368 | ||
e80d6a24 MG |
1369 | if (mtype < MIGRATE_TYPES) |
1370 | count[mtype]++; | |
467c996c MG |
1371 | } |
1372 | ||
1373 | /* Print counts */ | |
1374 | seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name); | |
1375 | for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) | |
1376 | seq_printf(m, "%12lu ", count[mtype]); | |
1377 | seq_putc(m, '\n'); | |
1378 | } | |
1379 | ||
f113e641 | 1380 | /* Print out the number of pageblocks for each migratetype */ |
467c996c MG |
1381 | static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg) |
1382 | { | |
1383 | int mtype; | |
1384 | pg_data_t *pgdat = (pg_data_t *)arg; | |
1385 | ||
1386 | seq_printf(m, "\n%-23s", "Number of blocks type "); | |
1387 | for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) | |
1388 | seq_printf(m, "%12s ", migratetype_names[mtype]); | |
1389 | seq_putc(m, '\n'); | |
727c080f VM |
1390 | walk_zones_in_node(m, pgdat, true, false, |
1391 | pagetypeinfo_showblockcount_print); | |
467c996c MG |
1392 | |
1393 | return 0; | |
1394 | } | |
1395 | ||
48c96a36 JK |
1396 | /* |
1397 | * Print out the number of pageblocks for each migratetype that contain pages | |
1398 | * of other types. This gives an indication of how well fallbacks are being | |
1399 | * contained by rmqueue_fallback(). It requires information from PAGE_OWNER | |
1400 | * to determine what is going on | |
1401 | */ | |
1402 | static void pagetypeinfo_showmixedcount(struct seq_file *m, pg_data_t *pgdat) | |
1403 | { | |
1404 | #ifdef CONFIG_PAGE_OWNER | |
1405 | int mtype; | |
1406 | ||
7dd80b8a | 1407 | if (!static_branch_unlikely(&page_owner_inited)) |
48c96a36 JK |
1408 | return; |
1409 | ||
1410 | drain_all_pages(NULL); | |
1411 | ||
1412 | seq_printf(m, "\n%-23s", "Number of mixed blocks "); | |
1413 | for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) | |
1414 | seq_printf(m, "%12s ", migratetype_names[mtype]); | |
1415 | seq_putc(m, '\n'); | |
1416 | ||
727c080f VM |
1417 | walk_zones_in_node(m, pgdat, true, true, |
1418 | pagetypeinfo_showmixedcount_print); | |
48c96a36 JK |
1419 | #endif /* CONFIG_PAGE_OWNER */ |
1420 | } | |
1421 | ||
467c996c MG |
1422 | /* |
1423 | * This prints out statistics in relation to grouping pages by mobility. | |
1424 | * It is expensive to collect so do not constantly read the file. | |
1425 | */ | |
1426 | static int pagetypeinfo_show(struct seq_file *m, void *arg) | |
1427 | { | |
1428 | pg_data_t *pgdat = (pg_data_t *)arg; | |
1429 | ||
41b25a37 | 1430 | /* check memoryless node */ |
a47b53c5 | 1431 | if (!node_state(pgdat->node_id, N_MEMORY)) |
41b25a37 KM |
1432 | return 0; |
1433 | ||
467c996c MG |
1434 | seq_printf(m, "Page block order: %d\n", pageblock_order); |
1435 | seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages); | |
1436 | seq_putc(m, '\n'); | |
1437 | pagetypeinfo_showfree(m, pgdat); | |
1438 | pagetypeinfo_showblockcount(m, pgdat); | |
48c96a36 | 1439 | pagetypeinfo_showmixedcount(m, pgdat); |
467c996c | 1440 | |
f6ac2354 CL |
1441 | return 0; |
1442 | } | |
1443 | ||
8f32f7e5 | 1444 | static const struct seq_operations fragmentation_op = { |
f6ac2354 CL |
1445 | .start = frag_start, |
1446 | .next = frag_next, | |
1447 | .stop = frag_stop, | |
1448 | .show = frag_show, | |
1449 | }; | |
1450 | ||
8f32f7e5 AD |
1451 | static int fragmentation_open(struct inode *inode, struct file *file) |
1452 | { | |
1453 | return seq_open(file, &fragmentation_op); | |
1454 | } | |
1455 | ||
9d85e15f | 1456 | static const struct file_operations buddyinfo_file_operations = { |
8f32f7e5 AD |
1457 | .open = fragmentation_open, |
1458 | .read = seq_read, | |
1459 | .llseek = seq_lseek, | |
1460 | .release = seq_release, | |
1461 | }; | |
1462 | ||
74e2e8e8 | 1463 | static const struct seq_operations pagetypeinfo_op = { |
467c996c MG |
1464 | .start = frag_start, |
1465 | .next = frag_next, | |
1466 | .stop = frag_stop, | |
1467 | .show = pagetypeinfo_show, | |
1468 | }; | |
1469 | ||
74e2e8e8 AD |
1470 | static int pagetypeinfo_open(struct inode *inode, struct file *file) |
1471 | { | |
1472 | return seq_open(file, &pagetypeinfo_op); | |
1473 | } | |
1474 | ||
9d85e15f | 1475 | static const struct file_operations pagetypeinfo_file_operations = { |
74e2e8e8 AD |
1476 | .open = pagetypeinfo_open, |
1477 | .read = seq_read, | |
1478 | .llseek = seq_lseek, | |
1479 | .release = seq_release, | |
1480 | }; | |
1481 | ||
e2ecc8a7 MG |
1482 | static bool is_zone_first_populated(pg_data_t *pgdat, struct zone *zone) |
1483 | { | |
1484 | int zid; | |
1485 | ||
1486 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1487 | struct zone *compare = &pgdat->node_zones[zid]; | |
1488 | ||
1489 | if (populated_zone(compare)) | |
1490 | return zone == compare; | |
1491 | } | |
1492 | ||
e2ecc8a7 MG |
1493 | return false; |
1494 | } | |
1495 | ||
467c996c MG |
1496 | static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat, |
1497 | struct zone *zone) | |
f6ac2354 | 1498 | { |
467c996c MG |
1499 | int i; |
1500 | seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name); | |
e2ecc8a7 MG |
1501 | if (is_zone_first_populated(pgdat, zone)) { |
1502 | seq_printf(m, "\n per-node stats"); | |
1503 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) { | |
1504 | seq_printf(m, "\n %-12s %lu", | |
3a321d2a KW |
1505 | vmstat_text[i + NR_VM_ZONE_STAT_ITEMS + |
1506 | NR_VM_NUMA_STAT_ITEMS], | |
e2ecc8a7 MG |
1507 | node_page_state(pgdat, i)); |
1508 | } | |
1509 | } | |
467c996c MG |
1510 | seq_printf(m, |
1511 | "\n pages free %lu" | |
1512 | "\n min %lu" | |
1513 | "\n low %lu" | |
1514 | "\n high %lu" | |
467c996c | 1515 | "\n spanned %lu" |
9feedc9d JL |
1516 | "\n present %lu" |
1517 | "\n managed %lu", | |
88f5acf8 | 1518 | zone_page_state(zone, NR_FREE_PAGES), |
41858966 MG |
1519 | min_wmark_pages(zone), |
1520 | low_wmark_pages(zone), | |
1521 | high_wmark_pages(zone), | |
467c996c | 1522 | zone->spanned_pages, |
9feedc9d JL |
1523 | zone->present_pages, |
1524 | zone->managed_pages); | |
467c996c | 1525 | |
467c996c | 1526 | seq_printf(m, |
3484b2de | 1527 | "\n protection: (%ld", |
467c996c MG |
1528 | zone->lowmem_reserve[0]); |
1529 | for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++) | |
3484b2de | 1530 | seq_printf(m, ", %ld", zone->lowmem_reserve[i]); |
7dfb8bf3 DR |
1531 | seq_putc(m, ')'); |
1532 | ||
1533 | /* If unpopulated, no other information is useful */ | |
1534 | if (!populated_zone(zone)) { | |
1535 | seq_putc(m, '\n'); | |
1536 | return; | |
1537 | } | |
1538 | ||
1539 | for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) | |
1540 | seq_printf(m, "\n %-12s %lu", vmstat_text[i], | |
1541 | zone_page_state(zone, i)); | |
1542 | ||
3a321d2a KW |
1543 | #ifdef CONFIG_NUMA |
1544 | for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++) | |
1545 | seq_printf(m, "\n %-12s %lu", | |
1546 | vmstat_text[i + NR_VM_ZONE_STAT_ITEMS], | |
1547 | zone_numa_state(zone, i)); | |
1548 | #endif | |
1549 | ||
7dfb8bf3 | 1550 | seq_printf(m, "\n pagesets"); |
467c996c MG |
1551 | for_each_online_cpu(i) { |
1552 | struct per_cpu_pageset *pageset; | |
467c996c | 1553 | |
99dcc3e5 | 1554 | pageset = per_cpu_ptr(zone->pageset, i); |
3dfa5721 CL |
1555 | seq_printf(m, |
1556 | "\n cpu: %i" | |
1557 | "\n count: %i" | |
1558 | "\n high: %i" | |
1559 | "\n batch: %i", | |
1560 | i, | |
1561 | pageset->pcp.count, | |
1562 | pageset->pcp.high, | |
1563 | pageset->pcp.batch); | |
df9ecaba | 1564 | #ifdef CONFIG_SMP |
467c996c MG |
1565 | seq_printf(m, "\n vm stats threshold: %d", |
1566 | pageset->stat_threshold); | |
df9ecaba | 1567 | #endif |
f6ac2354 | 1568 | } |
467c996c | 1569 | seq_printf(m, |
599d0c95 MG |
1570 | "\n node_unreclaimable: %u" |
1571 | "\n start_pfn: %lu" | |
1572 | "\n node_inactive_ratio: %u", | |
c73322d0 | 1573 | pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES, |
556adecb | 1574 | zone->zone_start_pfn, |
599d0c95 | 1575 | zone->zone_pgdat->inactive_ratio); |
467c996c MG |
1576 | seq_putc(m, '\n'); |
1577 | } | |
1578 | ||
1579 | /* | |
b2bd8598 DR |
1580 | * Output information about zones in @pgdat. All zones are printed regardless |
1581 | * of whether they are populated or not: lowmem_reserve_ratio operates on the | |
1582 | * set of all zones and userspace would not be aware of such zones if they are | |
1583 | * suppressed here (zoneinfo displays the effect of lowmem_reserve_ratio). | |
467c996c MG |
1584 | */ |
1585 | static int zoneinfo_show(struct seq_file *m, void *arg) | |
1586 | { | |
1587 | pg_data_t *pgdat = (pg_data_t *)arg; | |
727c080f | 1588 | walk_zones_in_node(m, pgdat, false, false, zoneinfo_show_print); |
f6ac2354 CL |
1589 | return 0; |
1590 | } | |
1591 | ||
5c9fe628 | 1592 | static const struct seq_operations zoneinfo_op = { |
f6ac2354 CL |
1593 | .start = frag_start, /* iterate over all zones. The same as in |
1594 | * fragmentation. */ | |
1595 | .next = frag_next, | |
1596 | .stop = frag_stop, | |
1597 | .show = zoneinfo_show, | |
1598 | }; | |
1599 | ||
5c9fe628 AD |
1600 | static int zoneinfo_open(struct inode *inode, struct file *file) |
1601 | { | |
1602 | return seq_open(file, &zoneinfo_op); | |
1603 | } | |
1604 | ||
9d85e15f | 1605 | static const struct file_operations zoneinfo_file_operations = { |
5c9fe628 AD |
1606 | .open = zoneinfo_open, |
1607 | .read = seq_read, | |
1608 | .llseek = seq_lseek, | |
1609 | .release = seq_release, | |
1610 | }; | |
1611 | ||
79da826a MR |
1612 | enum writeback_stat_item { |
1613 | NR_DIRTY_THRESHOLD, | |
1614 | NR_DIRTY_BG_THRESHOLD, | |
1615 | NR_VM_WRITEBACK_STAT_ITEMS, | |
1616 | }; | |
1617 | ||
f6ac2354 CL |
1618 | static void *vmstat_start(struct seq_file *m, loff_t *pos) |
1619 | { | |
2244b95a | 1620 | unsigned long *v; |
79da826a | 1621 | int i, stat_items_size; |
f6ac2354 CL |
1622 | |
1623 | if (*pos >= ARRAY_SIZE(vmstat_text)) | |
1624 | return NULL; | |
79da826a | 1625 | stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) + |
3a321d2a | 1626 | NR_VM_NUMA_STAT_ITEMS * sizeof(unsigned long) + |
75ef7184 | 1627 | NR_VM_NODE_STAT_ITEMS * sizeof(unsigned long) + |
79da826a | 1628 | NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long); |
f6ac2354 | 1629 | |
f8891e5e | 1630 | #ifdef CONFIG_VM_EVENT_COUNTERS |
79da826a | 1631 | stat_items_size += sizeof(struct vm_event_state); |
f8891e5e | 1632 | #endif |
79da826a MR |
1633 | |
1634 | v = kmalloc(stat_items_size, GFP_KERNEL); | |
2244b95a CL |
1635 | m->private = v; |
1636 | if (!v) | |
f6ac2354 | 1637 | return ERR_PTR(-ENOMEM); |
2244b95a | 1638 | for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) |
c41f012a | 1639 | v[i] = global_zone_page_state(i); |
79da826a MR |
1640 | v += NR_VM_ZONE_STAT_ITEMS; |
1641 | ||
3a321d2a KW |
1642 | #ifdef CONFIG_NUMA |
1643 | for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++) | |
1644 | v[i] = global_numa_state(i); | |
1645 | v += NR_VM_NUMA_STAT_ITEMS; | |
1646 | #endif | |
1647 | ||
75ef7184 MG |
1648 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
1649 | v[i] = global_node_page_state(i); | |
1650 | v += NR_VM_NODE_STAT_ITEMS; | |
1651 | ||
79da826a MR |
1652 | global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD, |
1653 | v + NR_DIRTY_THRESHOLD); | |
1654 | v += NR_VM_WRITEBACK_STAT_ITEMS; | |
1655 | ||
f8891e5e | 1656 | #ifdef CONFIG_VM_EVENT_COUNTERS |
79da826a MR |
1657 | all_vm_events(v); |
1658 | v[PGPGIN] /= 2; /* sectors -> kbytes */ | |
1659 | v[PGPGOUT] /= 2; | |
f8891e5e | 1660 | #endif |
ff8b16d7 | 1661 | return (unsigned long *)m->private + *pos; |
f6ac2354 CL |
1662 | } |
1663 | ||
1664 | static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos) | |
1665 | { | |
1666 | (*pos)++; | |
1667 | if (*pos >= ARRAY_SIZE(vmstat_text)) | |
1668 | return NULL; | |
1669 | return (unsigned long *)m->private + *pos; | |
1670 | } | |
1671 | ||
1672 | static int vmstat_show(struct seq_file *m, void *arg) | |
1673 | { | |
1674 | unsigned long *l = arg; | |
1675 | unsigned long off = l - (unsigned long *)m->private; | |
68ba0326 AD |
1676 | |
1677 | seq_puts(m, vmstat_text[off]); | |
75ba1d07 | 1678 | seq_put_decimal_ull(m, " ", *l); |
68ba0326 | 1679 | seq_putc(m, '\n'); |
f6ac2354 CL |
1680 | return 0; |
1681 | } | |
1682 | ||
1683 | static void vmstat_stop(struct seq_file *m, void *arg) | |
1684 | { | |
1685 | kfree(m->private); | |
1686 | m->private = NULL; | |
1687 | } | |
1688 | ||
b6aa44ab | 1689 | static const struct seq_operations vmstat_op = { |
f6ac2354 CL |
1690 | .start = vmstat_start, |
1691 | .next = vmstat_next, | |
1692 | .stop = vmstat_stop, | |
1693 | .show = vmstat_show, | |
1694 | }; | |
1695 | ||
b6aa44ab AD |
1696 | static int vmstat_open(struct inode *inode, struct file *file) |
1697 | { | |
1698 | return seq_open(file, &vmstat_op); | |
1699 | } | |
1700 | ||
9d85e15f | 1701 | static const struct file_operations vmstat_file_operations = { |
b6aa44ab AD |
1702 | .open = vmstat_open, |
1703 | .read = seq_read, | |
1704 | .llseek = seq_lseek, | |
1705 | .release = seq_release, | |
1706 | }; | |
f6ac2354 CL |
1707 | #endif /* CONFIG_PROC_FS */ |
1708 | ||
df9ecaba | 1709 | #ifdef CONFIG_SMP |
d1187ed2 | 1710 | static DEFINE_PER_CPU(struct delayed_work, vmstat_work); |
77461ab3 | 1711 | int sysctl_stat_interval __read_mostly = HZ; |
d1187ed2 | 1712 | |
52b6f46b HD |
1713 | #ifdef CONFIG_PROC_FS |
1714 | static void refresh_vm_stats(struct work_struct *work) | |
1715 | { | |
1716 | refresh_cpu_vm_stats(true); | |
1717 | } | |
1718 | ||
1719 | int vmstat_refresh(struct ctl_table *table, int write, | |
1720 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
1721 | { | |
1722 | long val; | |
1723 | int err; | |
1724 | int i; | |
1725 | ||
1726 | /* | |
1727 | * The regular update, every sysctl_stat_interval, may come later | |
1728 | * than expected: leaving a significant amount in per_cpu buckets. | |
1729 | * This is particularly misleading when checking a quantity of HUGE | |
1730 | * pages, immediately after running a test. /proc/sys/vm/stat_refresh, | |
1731 | * which can equally be echo'ed to or cat'ted from (by root), | |
1732 | * can be used to update the stats just before reading them. | |
1733 | * | |
c41f012a | 1734 | * Oh, and since global_zone_page_state() etc. are so careful to hide |
52b6f46b HD |
1735 | * transiently negative values, report an error here if any of |
1736 | * the stats is negative, so we know to go looking for imbalance. | |
1737 | */ | |
1738 | err = schedule_on_each_cpu(refresh_vm_stats); | |
1739 | if (err) | |
1740 | return err; | |
1741 | for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) { | |
75ef7184 | 1742 | val = atomic_long_read(&vm_zone_stat[i]); |
52b6f46b | 1743 | if (val < 0) { |
c822f622 JW |
1744 | pr_warn("%s: %s %ld\n", |
1745 | __func__, vmstat_text[i], val); | |
1746 | err = -EINVAL; | |
52b6f46b HD |
1747 | } |
1748 | } | |
3a321d2a KW |
1749 | #ifdef CONFIG_NUMA |
1750 | for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++) { | |
1751 | val = atomic_long_read(&vm_numa_stat[i]); | |
1752 | if (val < 0) { | |
1753 | pr_warn("%s: %s %ld\n", | |
1754 | __func__, vmstat_text[i + NR_VM_ZONE_STAT_ITEMS], val); | |
1755 | err = -EINVAL; | |
1756 | } | |
1757 | } | |
1758 | #endif | |
52b6f46b HD |
1759 | if (err) |
1760 | return err; | |
1761 | if (write) | |
1762 | *ppos += *lenp; | |
1763 | else | |
1764 | *lenp = 0; | |
1765 | return 0; | |
1766 | } | |
1767 | #endif /* CONFIG_PROC_FS */ | |
1768 | ||
d1187ed2 CL |
1769 | static void vmstat_update(struct work_struct *w) |
1770 | { | |
0eb77e98 | 1771 | if (refresh_cpu_vm_stats(true)) { |
7cc36bbd CL |
1772 | /* |
1773 | * Counters were updated so we expect more updates | |
1774 | * to occur in the future. Keep on running the | |
1775 | * update worker thread. | |
1776 | */ | |
ce612879 | 1777 | queue_delayed_work_on(smp_processor_id(), mm_percpu_wq, |
f01f17d3 MH |
1778 | this_cpu_ptr(&vmstat_work), |
1779 | round_jiffies_relative(sysctl_stat_interval)); | |
7cc36bbd CL |
1780 | } |
1781 | } | |
1782 | ||
0eb77e98 CL |
1783 | /* |
1784 | * Switch off vmstat processing and then fold all the remaining differentials | |
1785 | * until the diffs stay at zero. The function is used by NOHZ and can only be | |
1786 | * invoked when tick processing is not active. | |
1787 | */ | |
7cc36bbd CL |
1788 | /* |
1789 | * Check if the diffs for a certain cpu indicate that | |
1790 | * an update is needed. | |
1791 | */ | |
1792 | static bool need_update(int cpu) | |
1793 | { | |
1794 | struct zone *zone; | |
1795 | ||
1796 | for_each_populated_zone(zone) { | |
1797 | struct per_cpu_pageset *p = per_cpu_ptr(zone->pageset, cpu); | |
1798 | ||
1799 | BUILD_BUG_ON(sizeof(p->vm_stat_diff[0]) != 1); | |
3a321d2a KW |
1800 | #ifdef CONFIG_NUMA |
1801 | BUILD_BUG_ON(sizeof(p->vm_numa_stat_diff[0]) != 1); | |
1802 | #endif | |
7cc36bbd CL |
1803 | /* |
1804 | * The fast way of checking if there are any vmstat diffs. | |
1805 | * This works because the diffs are byte sized items. | |
1806 | */ | |
1807 | if (memchr_inv(p->vm_stat_diff, 0, NR_VM_ZONE_STAT_ITEMS)) | |
1808 | return true; | |
3a321d2a KW |
1809 | #ifdef CONFIG_NUMA |
1810 | if (memchr_inv(p->vm_numa_stat_diff, 0, NR_VM_NUMA_STAT_ITEMS)) | |
1811 | return true; | |
1812 | #endif | |
7cc36bbd CL |
1813 | } |
1814 | return false; | |
1815 | } | |
1816 | ||
7b8da4c7 CL |
1817 | /* |
1818 | * Switch off vmstat processing and then fold all the remaining differentials | |
1819 | * until the diffs stay at zero. The function is used by NOHZ and can only be | |
1820 | * invoked when tick processing is not active. | |
1821 | */ | |
f01f17d3 MH |
1822 | void quiet_vmstat(void) |
1823 | { | |
1824 | if (system_state != SYSTEM_RUNNING) | |
1825 | return; | |
1826 | ||
7b8da4c7 | 1827 | if (!delayed_work_pending(this_cpu_ptr(&vmstat_work))) |
f01f17d3 MH |
1828 | return; |
1829 | ||
1830 | if (!need_update(smp_processor_id())) | |
1831 | return; | |
1832 | ||
1833 | /* | |
1834 | * Just refresh counters and do not care about the pending delayed | |
1835 | * vmstat_update. It doesn't fire that often to matter and canceling | |
1836 | * it would be too expensive from this path. | |
1837 | * vmstat_shepherd will take care about that for us. | |
1838 | */ | |
1839 | refresh_cpu_vm_stats(false); | |
1840 | } | |
1841 | ||
7cc36bbd CL |
1842 | /* |
1843 | * Shepherd worker thread that checks the | |
1844 | * differentials of processors that have their worker | |
1845 | * threads for vm statistics updates disabled because of | |
1846 | * inactivity. | |
1847 | */ | |
1848 | static void vmstat_shepherd(struct work_struct *w); | |
1849 | ||
0eb77e98 | 1850 | static DECLARE_DEFERRABLE_WORK(shepherd, vmstat_shepherd); |
7cc36bbd CL |
1851 | |
1852 | static void vmstat_shepherd(struct work_struct *w) | |
1853 | { | |
1854 | int cpu; | |
1855 | ||
1856 | get_online_cpus(); | |
1857 | /* Check processors whose vmstat worker threads have been disabled */ | |
7b8da4c7 | 1858 | for_each_online_cpu(cpu) { |
f01f17d3 | 1859 | struct delayed_work *dw = &per_cpu(vmstat_work, cpu); |
7cc36bbd | 1860 | |
7b8da4c7 | 1861 | if (!delayed_work_pending(dw) && need_update(cpu)) |
ce612879 | 1862 | queue_delayed_work_on(cpu, mm_percpu_wq, dw, 0); |
f01f17d3 | 1863 | } |
7cc36bbd CL |
1864 | put_online_cpus(); |
1865 | ||
1866 | schedule_delayed_work(&shepherd, | |
98f4ebb2 | 1867 | round_jiffies_relative(sysctl_stat_interval)); |
d1187ed2 CL |
1868 | } |
1869 | ||
7cc36bbd | 1870 | static void __init start_shepherd_timer(void) |
d1187ed2 | 1871 | { |
7cc36bbd CL |
1872 | int cpu; |
1873 | ||
1874 | for_each_possible_cpu(cpu) | |
ccde8bd4 | 1875 | INIT_DEFERRABLE_WORK(per_cpu_ptr(&vmstat_work, cpu), |
7cc36bbd CL |
1876 | vmstat_update); |
1877 | ||
7cc36bbd CL |
1878 | schedule_delayed_work(&shepherd, |
1879 | round_jiffies_relative(sysctl_stat_interval)); | |
d1187ed2 CL |
1880 | } |
1881 | ||
03e86dba TC |
1882 | static void __init init_cpu_node_state(void) |
1883 | { | |
4c501327 | 1884 | int node; |
03e86dba | 1885 | |
4c501327 SAS |
1886 | for_each_online_node(node) { |
1887 | if (cpumask_weight(cpumask_of_node(node)) > 0) | |
1888 | node_set_state(node, N_CPU); | |
1889 | } | |
03e86dba TC |
1890 | } |
1891 | ||
5438da97 SAS |
1892 | static int vmstat_cpu_online(unsigned int cpu) |
1893 | { | |
1894 | refresh_zone_stat_thresholds(); | |
1895 | node_set_state(cpu_to_node(cpu), N_CPU); | |
1896 | return 0; | |
1897 | } | |
1898 | ||
1899 | static int vmstat_cpu_down_prep(unsigned int cpu) | |
1900 | { | |
1901 | cancel_delayed_work_sync(&per_cpu(vmstat_work, cpu)); | |
1902 | return 0; | |
1903 | } | |
1904 | ||
1905 | static int vmstat_cpu_dead(unsigned int cpu) | |
807a1bd2 | 1906 | { |
4c501327 | 1907 | const struct cpumask *node_cpus; |
5438da97 | 1908 | int node; |
807a1bd2 | 1909 | |
5438da97 SAS |
1910 | node = cpu_to_node(cpu); |
1911 | ||
1912 | refresh_zone_stat_thresholds(); | |
4c501327 SAS |
1913 | node_cpus = cpumask_of_node(node); |
1914 | if (cpumask_weight(node_cpus) > 0) | |
5438da97 | 1915 | return 0; |
807a1bd2 TK |
1916 | |
1917 | node_clear_state(node, N_CPU); | |
5438da97 | 1918 | return 0; |
807a1bd2 TK |
1919 | } |
1920 | ||
8f32f7e5 | 1921 | #endif |
df9ecaba | 1922 | |
ce612879 MH |
1923 | struct workqueue_struct *mm_percpu_wq; |
1924 | ||
597b7305 | 1925 | void __init init_mm_internals(void) |
df9ecaba | 1926 | { |
ce612879 | 1927 | int ret __maybe_unused; |
5438da97 | 1928 | |
80d136e1 | 1929 | mm_percpu_wq = alloc_workqueue("mm_percpu_wq", WQ_MEM_RECLAIM, 0); |
ce612879 MH |
1930 | |
1931 | #ifdef CONFIG_SMP | |
5438da97 SAS |
1932 | ret = cpuhp_setup_state_nocalls(CPUHP_MM_VMSTAT_DEAD, "mm/vmstat:dead", |
1933 | NULL, vmstat_cpu_dead); | |
1934 | if (ret < 0) | |
1935 | pr_err("vmstat: failed to register 'dead' hotplug state\n"); | |
1936 | ||
1937 | ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "mm/vmstat:online", | |
1938 | vmstat_cpu_online, | |
1939 | vmstat_cpu_down_prep); | |
1940 | if (ret < 0) | |
1941 | pr_err("vmstat: failed to register 'online' hotplug state\n"); | |
1942 | ||
1943 | get_online_cpus(); | |
03e86dba | 1944 | init_cpu_node_state(); |
5438da97 | 1945 | put_online_cpus(); |
d1187ed2 | 1946 | |
7cc36bbd | 1947 | start_shepherd_timer(); |
8f32f7e5 AD |
1948 | #endif |
1949 | #ifdef CONFIG_PROC_FS | |
9d85e15f AK |
1950 | proc_create("buddyinfo", 0444, NULL, &buddyinfo_file_operations); |
1951 | proc_create("pagetypeinfo", 0444, NULL, &pagetypeinfo_file_operations); | |
1952 | proc_create("vmstat", 0444, NULL, &vmstat_file_operations); | |
1953 | proc_create("zoneinfo", 0444, NULL, &zoneinfo_file_operations); | |
8f32f7e5 | 1954 | #endif |
df9ecaba | 1955 | } |
d7a5752c MG |
1956 | |
1957 | #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION) | |
d7a5752c MG |
1958 | |
1959 | /* | |
1960 | * Return an index indicating how much of the available free memory is | |
1961 | * unusable for an allocation of the requested size. | |
1962 | */ | |
1963 | static int unusable_free_index(unsigned int order, | |
1964 | struct contig_page_info *info) | |
1965 | { | |
1966 | /* No free memory is interpreted as all free memory is unusable */ | |
1967 | if (info->free_pages == 0) | |
1968 | return 1000; | |
1969 | ||
1970 | /* | |
1971 | * Index should be a value between 0 and 1. Return a value to 3 | |
1972 | * decimal places. | |
1973 | * | |
1974 | * 0 => no fragmentation | |
1975 | * 1 => high fragmentation | |
1976 | */ | |
1977 | return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages); | |
1978 | ||
1979 | } | |
1980 | ||
1981 | static void unusable_show_print(struct seq_file *m, | |
1982 | pg_data_t *pgdat, struct zone *zone) | |
1983 | { | |
1984 | unsigned int order; | |
1985 | int index; | |
1986 | struct contig_page_info info; | |
1987 | ||
1988 | seq_printf(m, "Node %d, zone %8s ", | |
1989 | pgdat->node_id, | |
1990 | zone->name); | |
1991 | for (order = 0; order < MAX_ORDER; ++order) { | |
1992 | fill_contig_page_info(zone, order, &info); | |
1993 | index = unusable_free_index(order, &info); | |
1994 | seq_printf(m, "%d.%03d ", index / 1000, index % 1000); | |
1995 | } | |
1996 | ||
1997 | seq_putc(m, '\n'); | |
1998 | } | |
1999 | ||
2000 | /* | |
2001 | * Display unusable free space index | |
2002 | * | |
2003 | * The unusable free space index measures how much of the available free | |
2004 | * memory cannot be used to satisfy an allocation of a given size and is a | |
2005 | * value between 0 and 1. The higher the value, the more of free memory is | |
2006 | * unusable and by implication, the worse the external fragmentation is. This | |
2007 | * can be expressed as a percentage by multiplying by 100. | |
2008 | */ | |
2009 | static int unusable_show(struct seq_file *m, void *arg) | |
2010 | { | |
2011 | pg_data_t *pgdat = (pg_data_t *)arg; | |
2012 | ||
2013 | /* check memoryless node */ | |
a47b53c5 | 2014 | if (!node_state(pgdat->node_id, N_MEMORY)) |
d7a5752c MG |
2015 | return 0; |
2016 | ||
727c080f | 2017 | walk_zones_in_node(m, pgdat, true, false, unusable_show_print); |
d7a5752c MG |
2018 | |
2019 | return 0; | |
2020 | } | |
2021 | ||
2022 | static const struct seq_operations unusable_op = { | |
2023 | .start = frag_start, | |
2024 | .next = frag_next, | |
2025 | .stop = frag_stop, | |
2026 | .show = unusable_show, | |
2027 | }; | |
2028 | ||
2029 | static int unusable_open(struct inode *inode, struct file *file) | |
2030 | { | |
2031 | return seq_open(file, &unusable_op); | |
2032 | } | |
2033 | ||
2034 | static const struct file_operations unusable_file_ops = { | |
2035 | .open = unusable_open, | |
2036 | .read = seq_read, | |
2037 | .llseek = seq_lseek, | |
2038 | .release = seq_release, | |
2039 | }; | |
2040 | ||
f1a5ab12 MG |
2041 | static void extfrag_show_print(struct seq_file *m, |
2042 | pg_data_t *pgdat, struct zone *zone) | |
2043 | { | |
2044 | unsigned int order; | |
2045 | int index; | |
2046 | ||
2047 | /* Alloc on stack as interrupts are disabled for zone walk */ | |
2048 | struct contig_page_info info; | |
2049 | ||
2050 | seq_printf(m, "Node %d, zone %8s ", | |
2051 | pgdat->node_id, | |
2052 | zone->name); | |
2053 | for (order = 0; order < MAX_ORDER; ++order) { | |
2054 | fill_contig_page_info(zone, order, &info); | |
56de7263 | 2055 | index = __fragmentation_index(order, &info); |
f1a5ab12 MG |
2056 | seq_printf(m, "%d.%03d ", index / 1000, index % 1000); |
2057 | } | |
2058 | ||
2059 | seq_putc(m, '\n'); | |
2060 | } | |
2061 | ||
2062 | /* | |
2063 | * Display fragmentation index for orders that allocations would fail for | |
2064 | */ | |
2065 | static int extfrag_show(struct seq_file *m, void *arg) | |
2066 | { | |
2067 | pg_data_t *pgdat = (pg_data_t *)arg; | |
2068 | ||
727c080f | 2069 | walk_zones_in_node(m, pgdat, true, false, extfrag_show_print); |
f1a5ab12 MG |
2070 | |
2071 | return 0; | |
2072 | } | |
2073 | ||
2074 | static const struct seq_operations extfrag_op = { | |
2075 | .start = frag_start, | |
2076 | .next = frag_next, | |
2077 | .stop = frag_stop, | |
2078 | .show = extfrag_show, | |
2079 | }; | |
2080 | ||
2081 | static int extfrag_open(struct inode *inode, struct file *file) | |
2082 | { | |
2083 | return seq_open(file, &extfrag_op); | |
2084 | } | |
2085 | ||
2086 | static const struct file_operations extfrag_file_ops = { | |
2087 | .open = extfrag_open, | |
2088 | .read = seq_read, | |
2089 | .llseek = seq_lseek, | |
2090 | .release = seq_release, | |
2091 | }; | |
2092 | ||
d7a5752c MG |
2093 | static int __init extfrag_debug_init(void) |
2094 | { | |
bde8bd8a S |
2095 | struct dentry *extfrag_debug_root; |
2096 | ||
d7a5752c MG |
2097 | extfrag_debug_root = debugfs_create_dir("extfrag", NULL); |
2098 | if (!extfrag_debug_root) | |
2099 | return -ENOMEM; | |
2100 | ||
2101 | if (!debugfs_create_file("unusable_index", 0444, | |
2102 | extfrag_debug_root, NULL, &unusable_file_ops)) | |
bde8bd8a | 2103 | goto fail; |
d7a5752c | 2104 | |
f1a5ab12 MG |
2105 | if (!debugfs_create_file("extfrag_index", 0444, |
2106 | extfrag_debug_root, NULL, &extfrag_file_ops)) | |
bde8bd8a | 2107 | goto fail; |
f1a5ab12 | 2108 | |
d7a5752c | 2109 | return 0; |
bde8bd8a S |
2110 | fail: |
2111 | debugfs_remove_recursive(extfrag_debug_root); | |
2112 | return -ENOMEM; | |
d7a5752c MG |
2113 | } |
2114 | ||
2115 | module_init(extfrag_debug_init); | |
2116 | #endif |