forked from iovisor/bcc
-
Notifications
You must be signed in to change notification settings - Fork 2
/
compactsnoop.py
executable file
·408 lines (361 loc) · 11.1 KB
/
compactsnoop.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
#!/usr/bin/env python
# @lint-avoid-python-3-compatibility-imports
#
# compactsnoop Trace compact zone and print details including issuing PID.
# For Linux, uses BCC, eBPF.
#
# This uses in-kernel eBPF maps to cache process details (PID and comm) by
# compact zone begin, as well as a starting timestamp for calculating
# latency.
#
# Copyright (c) 2019 Wenbo Zhang
# Licensed under the Apache License, Version 2.0 (the "License")
#
# 11-NOV-2019 Wenbo Zhang Created this.
from __future__ import print_function
from bcc import BPF
import argparse
import platform
from datetime import datetime, timedelta
import sys
# arguments
examples = """examples:
./compactsnoop # trace all compact stall
./compactsnoop -T # include timestamps
./compactsnoop -d 10 # trace for 10 seconds only
./compactsnoop -K # output kernel stack trace
./compactsnoop -e # show extended fields
"""
parser = argparse.ArgumentParser(
description="Trace compact zone",
formatter_class=argparse.RawDescriptionHelpFormatter,
epilog=examples,
)
parser.add_argument("-T", "--timestamp", action="store_true",
help="include timestamp on output")
parser.add_argument("-p", "--pid", help="trace this PID only")
parser.add_argument("-d", "--duration",
help="total duration of trace in seconds")
parser.add_argument("-K", "--kernel-stack", action="store_true",
help="output kernel stack trace")
parser.add_argument("-e", "--extended_fields", action="store_true",
help="show system memory state")
parser.add_argument("--ebpf", action="store_true", help=argparse.SUPPRESS)
args = parser.parse_args()
debug = 0
if args.duration:
args.duration = timedelta(seconds=int(args.duration))
NO_EXTENDED = """
#ifdef EXTNEDED_FIELDS
#undef EXTNEDED_FIELDS
#endif
"""
EXTENDED = """
#define EXTNEDED_FIELDS 1
"""
bpf_text = """
#include <uapi/linux/ptrace.h>
#include <linux/sched.h>
#include <linux/mmzone.h>
#include <linux/compaction.h>
struct val_t {
int nid;
int idx;
int order;
int sync;
#ifdef EXTNEDED_FIELDS
int fragindex;
int low;
int min;
int high;
int free;
#endif
u64 ts; // compaction begin time
};
struct data_t {
u32 pid;
u32 tid;
int nid;
int idx;
int order;
u64 delta;
u64 ts; // compaction end time
int sync;
#ifdef EXTNEDED_FIELDS
int fragindex;
int low;
int min;
int high;
int free;
#endif
int status;
int stack_id;
char comm[TASK_COMM_LEN];
};
BPF_HASH(start, u64, struct val_t);
BPF_PERF_OUTPUT(events);
BPF_STACK_TRACE(stack_traces, 2048);
#ifdef CONFIG_NUMA
static inline int zone_to_nid_(struct zone *zone)
{
int node;
bpf_probe_read_kernel(&node, sizeof(node), &zone->node);
return node;
}
#else
static inline int zone_to_nid_(struct zone *zone)
{
return 0;
}
#endif
// #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
static inline int zone_idx_(struct zone *zone)
{
struct pglist_data *zone_pgdat = NULL;
bpf_probe_read_kernel(&zone_pgdat, sizeof(zone_pgdat), &zone->zone_pgdat);
return ((u64)zone - (u64)zone_pgdat->node_zones)/sizeof(struct zone);
}
#ifdef EXTNEDED_FIELDS
static inline void get_all_wmark_pages(struct zone *zone, struct val_t *valp)
{
u64 _watermark[NR_WMARK] = {};
u64 watermark_boost = 0;
bpf_probe_read_kernel(&_watermark, sizeof(_watermark), &zone->_watermark);
bpf_probe_read_kernel(&watermark_boost, sizeof(watermark_boost),
&zone->watermark_boost);
valp->min = _watermark[WMARK_MIN] + watermark_boost;
valp->low = _watermark[WMARK_LOW] + watermark_boost;
valp->high = _watermark[WMARK_HIGH] + watermark_boost;
bpf_probe_read_kernel(&valp->free, sizeof(valp->free),
&zone->vm_stat[NR_FREE_PAGES]);
}
#endif
static inline void submit_event(void *ctx, int status)
{
struct data_t data = {};
u64 ts = bpf_ktime_get_ns();
u64 id = bpf_get_current_pid_tgid();
struct val_t *valp = start.lookup(&id);
if (valp == NULL) {
// missed entry
return;
}
data.delta = ts - valp->ts;
data.ts = ts / 1000;
data.pid = id >> 32;
data.tid = id;
bpf_get_current_comm(&data.comm, sizeof(data.comm));
data.nid = valp->nid;
data.idx = valp->idx;
data.order = valp->order;
data.sync = valp->sync;
#ifdef EXTNEDED_FIELDS
data.fragindex = valp->fragindex;
data.min = valp->min;
data.low = valp->low;
data.high = valp->high;
data.free = valp->free;
#endif
data.status = status;
data.stack_id = stack_traces.get_stackid(ctx, 0);
events.perf_submit(ctx, &data, sizeof(data));
start.delete(&id);
}
#ifdef EXTNEDED_FIELDS
int trace_fragmentation_index_return(struct pt_regs *ctx)
{
struct val_t val = { };
int ret = PT_REGS_RC(ctx);
u64 id = bpf_get_current_pid_tgid();
PID_FILTER
val.fragindex = ret;
start.update(&id, &val);
return 0;
}
#endif
static inline void fill_compact_info(struct val_t *valp,
struct zone *zone,
int order)
{
valp->nid = zone_to_nid_(zone);
valp->idx = zone_idx_(zone);
valp->order = order;
}
RAW_TRACEPOINT_PROBE(mm_compaction_suitable)
{
// TP_PROTO(struct zone *zone, int order, int ret)
struct zone *zone = (struct zone *)ctx->args[0];
struct val_t val = { };
int order = (int)ctx->args[1];
int ret = (int)ctx->args[2];
u64 id;
if(ret != COMPACT_CONTINUE)
return 0;
id = bpf_get_current_pid_tgid();
PID_FILTER
#ifdef EXTNEDED_FIELDS
struct val_t *valp = start.lookup(&id);
if (valp == NULL) {
// missed entry or order <= PAGE_ALLOC_COSTLY_ORDER, eg:
// manual trigger echo 1 > /proc/sys/vm/compact_memory
val.fragindex = -1000;
valp = &val;
}
fill_compact_info(valp, zone, order);
get_all_wmark_pages(zone, valp);
start.update(&id, valp);
#else
fill_compact_info(&val, zone, order);
start.update(&id, &val);
#endif
return 0;
}
TRACEPOINT_PROBE(compaction, mm_compaction_begin)
{
bool sync = args->sync;
u64 id = bpf_get_current_pid_tgid();
struct val_t *valp = start.lookup(&id);
if (valp == NULL) {
// missed entry
return 0;
}
valp->ts = bpf_ktime_get_ns();
valp->sync = sync;
return 0;
}
TRACEPOINT_PROBE(compaction, mm_compaction_end)
{
submit_event(args, args->status);
return 0;
}
"""
if platform.machine() != 'x86_64' and platform.machine() != 'ppc64le':
print("""
Currently only support x86_64 and power servers, if you want
to use it on other platforms(including power embedded processors),
please refer include/linux/mmzone.h to modify zone_idex_to_str to
get the right zone type
""")
exit()
if args.extended_fields:
bpf_text = EXTENDED + bpf_text
else:
bpf_text = NO_EXTENDED + bpf_text
if args.pid:
bpf_text = bpf_text.replace("PID_FILTER",
"if (id >> 32 != %s) { return 0; }" % args.pid)
else:
bpf_text = bpf_text.replace("PID_FILTER", "")
if debug or args.ebpf:
print(bpf_text)
if args.ebpf:
exit()
# load BPF program
b = BPF(text=bpf_text)
if args.extended_fields:
b.attach_kretprobe(event="fragmentation_index",
fn_name="trace_fragmentation_index_return")
stack_traces = b.get_table("stack_traces")
initial_ts = 0
def zone_idx_to_str(idx):
# from include/linux/mmzone.h
# NOTICE: consider only x86_64 servers
zone_type = {
'x86_64':
{
0: "ZONE_DMA",
1: "ZONE_DMA32",
2: "ZONE_NORMAL"
},
# Zones in Power server only
'ppc64le':
{
0: "ZONE_NORMAL",
1: "ZONE_MOVABLE"
}
}
if idx in zone_type[platform.machine()]:
return zone_type[platform.machine()][idx]
else:
return str(idx)
def compact_result_to_str(status):
# from include/trace/evnets/mmflags.h
# from include/linux/compaction.h
compact_status = {
# COMPACT_NOT_SUITABLE_ZONE: For more detailed tracepoint
# output - internal to compaction
0: "not_suitable_zone",
# COMPACT_SKIPPED: compaction didn't start as it was not
# possible or direct reclaim was more suitable
1: "skipped",
# COMPACT_DEFERRED: compaction didn't start as it was
# deferred due to past failures
2: "deferred",
# COMPACT_NOT_SUITABLE_PAGE: For more detailed tracepoint
# output - internal to compaction
3: "no_suitable_page",
# COMPACT_CONTINUE: compaction should continue to another pageblock
4: "continue",
# COMPACT_COMPLETE: The full zone was compacted scanned but wasn't
# successful to compact suitable pages.
5: "complete",
# COMPACT_PARTIAL_SKIPPED: direct compaction has scanned part of the
# zone but wasn't successful to compact suitable pages.
6: "partial_skipped",
# COMPACT_CONTENDED: compaction terminated prematurely due to lock
# contentions
7: "contended",
# COMPACT_SUCCESS: direct compaction terminated after concluding
# that the allocation should now succeed
8: "success",
}
if status in compact_status:
return compact_status[status]
else:
return str(status)
# header
if args.timestamp:
print("%-14s" % ("TIME(s)"), end=" ")
print("%-14s %-6s %-4s %-12s %-5s %-7s" %
("COMM", "PID", "NODE", "ZONE", "ORDER", "MODE"), end=" ")
if args.extended_fields:
print("%-8s %-8s %-8s %-8s %-8s" %
("FRAGIDX", "MIN", "LOW", "HIGH", "FREE"), end=" ")
print("%9s %16s" % ("LAT(ms)", "STATUS"))
# process event
def print_event(cpu, data, size):
event = b["events"].event(data)
global initial_ts
if not initial_ts:
initial_ts = event.ts
if args.timestamp:
delta = event.ts - initial_ts
print("%-14.9f" % (float(delta) / 1000000), end=" ")
print("%-14.14s %-6s %-4s %-12s %-5s %-7s" % (
event.comm.decode("utf-8", "replace"),
event.pid,
event.nid,
zone_idx_to_str(event.idx),
event.order,
"SYNC" if event.sync else "ASYNC"), end=" ")
if args.extended_fields:
print("%-8.3f %-8s %-8s %-8s %-8s" % (
(float(event.fragindex) / 1000),
event.min, event.low, event.high, event.free
), end=" ")
print("%9.3f %16s" % (
float(event.delta) / 1000000, compact_result_to_str(event.status)))
if args.kernel_stack:
for addr in stack_traces.walk(event.stack_id):
sym = b.ksym(addr, show_offset=True)
print("\t%s" % sym)
print("")
sys.stdout.flush()
# loop with callback to print_event
b["events"].open_perf_buffer(print_event, page_cnt=64)
start_time = datetime.now()
while not args.duration or datetime.now() - start_time < args.duration:
try:
b.perf_buffer_poll()
except KeyboardInterrupt:
exit()