1#!/bin/bash
2# perf record tests (exclusive)
3# SPDX-License-Identifier: GPL-2.0
4
5set -e
6
7shelldir=$(dirname "$0")
8# shellcheck source=lib/waiting.sh
9. "${shelldir}"/lib/waiting.sh
10
11# shellcheck source=lib/perf_has_symbol.sh
12. "${shelldir}"/lib/perf_has_symbol.sh
13
14testsym="test_loop"
15testsym2="brstack"
16
17skip_test_missing_symbol ${testsym}
18skip_test_missing_symbol ${testsym2}
19
20err=0
21perfdata=$(mktemp /tmp/__perf_test.perf.data.XXXXX)
22script_output=$(mktemp /tmp/__perf_test.perf.data.XXXXX.script)
23testprog="perf test -w thloop"
24cpu_pmu_dir="/sys/bus/event_source/devices/cpu*"
25br_cntr_file="/caps/branch_counter_nr"
26br_cntr_output="branch stack counters"
27br_cntr_script_output="br_cntr: A"
28
29default_fd_limit=$(ulimit -Sn)
30# With option --threads=cpu the number of open file descriptors should be
31# equal to sum of:    nmb_cpus * nmb_events (2+dummy),
32#                     nmb_threads for perf.data.n (equal to nmb_cpus) and
33#                     2*nmb_cpus of pipes = 4*nmb_cpus (each pipe has 2 ends)
34# All together it needs 8*nmb_cpus file descriptors plus some are also used
35# outside of testing, thus raising the limit to 16*nmb_cpus
36min_fd_limit=$(($(getconf _NPROCESSORS_ONLN) * 16))
37
38cleanup() {
39  rm -f "${perfdata}"
40  rm -f "${perfdata}".old
41  rm -f "${script_output}"
42
43  trap - EXIT TERM INT
44}
45
46trap_cleanup() {
47  echo "Unexpected signal in ${FUNCNAME[1]}"
48  cleanup
49  exit 1
50}
51trap trap_cleanup EXIT TERM INT
52
53test_per_thread() {
54  echo "Basic --per-thread mode test"
55  if ! perf record -o /dev/null --quiet ${testprog} 2> /dev/null
56  then
57    echo "Per-thread record [Skipped event not supported]"
58    return
59  fi
60  if ! perf record --per-thread -o "${perfdata}" ${testprog} 2> /dev/null
61  then
62    echo "Per-thread record [Failed record]"
63    err=1
64    return
65  fi
66  if ! perf report -i "${perfdata}" -q | grep -q "${testsym}"
67  then
68    echo "Per-thread record [Failed missing output]"
69    err=1
70    return
71  fi
72
73  # run the test program in background (for 30 seconds)
74  ${testprog} 30 &
75  TESTPID=$!
76
77  rm -f "${perfdata}"
78
79  wait_for_threads ${TESTPID} 2
80  perf record -p "${TESTPID}" --per-thread -o "${perfdata}" sleep 1 2> /dev/null
81  kill ${TESTPID}
82
83  if [ ! -e "${perfdata}" ]
84  then
85    echo "Per-thread record [Failed record -p]"
86    err=1
87    return
88  fi
89  if ! perf report -i "${perfdata}" -q | grep -q "${testsym}"
90  then
91    echo "Per-thread record [Failed -p missing output]"
92    err=1
93    return
94  fi
95
96  echo "Basic --per-thread mode test [Success]"
97}
98
99test_register_capture() {
100  echo "Register capture test"
101  if ! perf list pmu | grep -q 'br_inst_retired.near_call'
102  then
103    echo "Register capture test [Skipped missing event]"
104    return
105  fi
106  if ! perf record --intr-regs=\? 2>&1 | grep -q 'available registers: AX BX CX DX SI DI BP SP IP FLAGS CS SS R8 R9 R10 R11 R12 R13 R14 R15'
107  then
108    echo "Register capture test [Skipped missing registers]"
109    return
110  fi
111  if ! perf record -o - --intr-regs=di,r8,dx,cx -e br_inst_retired.near_call \
112    -c 1000 --per-thread ${testprog} 2> /dev/null \
113    | perf script -F ip,sym,iregs -i - 2> /dev/null \
114    | grep -q "DI:"
115  then
116    echo "Register capture test [Failed missing output]"
117    err=1
118    return
119  fi
120  echo "Register capture test [Success]"
121}
122
123test_system_wide() {
124  echo "Basic --system-wide mode test"
125  if ! perf record -aB --synth=no -o "${perfdata}" ${testprog} 2> /dev/null
126  then
127    echo "System-wide record [Skipped not supported]"
128    return
129  fi
130  if ! perf report -i "${perfdata}" -q | grep -q "${testsym}"
131  then
132    echo "System-wide record [Failed missing output]"
133    err=1
134    return
135  fi
136  if ! perf record -aB --synth=no -e cpu-clock,cs --threads=cpu \
137    -o "${perfdata}" ${testprog} 2> /dev/null
138  then
139    echo "System-wide record [Failed record --threads option]"
140    err=1
141    return
142  fi
143  if ! perf report -i "${perfdata}" -q | grep -q "${testsym}"
144  then
145    echo "System-wide record [Failed --threads missing output]"
146    err=1
147    return
148  fi
149  echo "Basic --system-wide mode test [Success]"
150}
151
152test_workload() {
153  echo "Basic target workload test"
154  if ! perf record -o "${perfdata}" ${testprog} 2> /dev/null
155  then
156    echo "Workload record [Failed record]"
157    err=1
158    return
159  fi
160  if ! perf report -i "${perfdata}" -q | grep -q "${testsym}"
161  then
162    echo "Workload record [Failed missing output]"
163    err=1
164    return
165  fi
166  if ! perf record -e cpu-clock,cs --threads=package \
167    -o "${perfdata}" ${testprog} 2> /dev/null
168  then
169    echo "Workload record [Failed record --threads option]"
170    err=1
171    return
172  fi
173  if ! perf report -i "${perfdata}" -q | grep -q "${testsym}"
174  then
175    echo "Workload record [Failed --threads missing output]"
176    err=1
177    return
178  fi
179  echo "Basic target workload test [Success]"
180}
181
182test_branch_counter() {
183  echo "Branch counter test"
184  # Check if the branch counter feature is supported
185  for dir in $cpu_pmu_dir
186  do
187    if [ ! -e "$dir$br_cntr_file" ]
188    then
189      echo "branch counter feature not supported on all core PMUs ($dir) [Skipped]"
190      return
191    fi
192  done
193  if ! perf record -o "${perfdata}" -e "{branches:p,instructions}" -j any,counter ${testprog} 2> /dev/null
194  then
195    echo "Branch counter record test [Failed record]"
196    err=1
197    return
198  fi
199  if ! perf report -i "${perfdata}" -D -q | grep -q "$br_cntr_output"
200  then
201    echo "Branch counter report test [Failed missing output]"
202    err=1
203    return
204  fi
205  if ! perf script -i "${perfdata}" -F +brstackinsn,+brcntr | grep -q "$br_cntr_script_output"
206  then
207    echo " Branch counter script test [Failed missing output]"
208    err=1
209    return
210  fi
211  echo "Branch counter test [Success]"
212}
213
214test_cgroup() {
215  echo "Cgroup sampling test"
216  if ! perf record -aB --synth=cgroup --all-cgroups -o "${perfdata}" ${testprog} 2> /dev/null
217  then
218    echo "Cgroup sampling [Skipped not supported]"
219    return
220  fi
221  if ! perf report -i "${perfdata}" -D | grep -q "CGROUP"
222  then
223    echo "Cgroup sampling [Failed missing output]"
224    err=1
225    return
226  fi
227  if ! perf script -i "${perfdata}" -F cgroup | grep -q -v "unknown"
228  then
229    echo "Cgroup sampling [Failed cannot resolve cgroup names]"
230    err=1
231    return
232  fi
233  echo "Cgroup sampling test [Success]"
234}
235
236test_uid() {
237  echo "Uid sampling test"
238  if ! perf record -aB --synth=no --uid "$(id -u)" -o "${perfdata}" ${testprog} \
239    > "${script_output}" 2>&1
240  then
241    if grep -q "libbpf.*EPERM" "${script_output}"
242    then
243      echo "Uid sampling [Skipped permissions]"
244      return
245    else
246      echo "Uid sampling [Failed to record]"
247      err=1
248      # cat "${script_output}"
249      return
250    fi
251  fi
252  if ! perf report -i "${perfdata}" -q | grep -q "${testsym}"
253  then
254    echo "Uid sampling [Failed missing output]"
255    err=1
256    return
257  fi
258  echo "Uid sampling test [Success]"
259}
260
261test_leader_sampling() {
262  echo "Basic leader sampling test"
263  if ! perf record -o "${perfdata}" -e "{cycles,cycles}:Su" -- \
264    perf test -w brstack 2> /dev/null
265  then
266    echo "Leader sampling [Failed record]"
267    err=1
268    return
269  fi
270  perf script -i "${perfdata}" | grep brstack > $script_output
271  # Check if the two instruction counts are equal in each record.
272  # However, the throttling code doesn't consider event grouping. During throttling, only the
273  # leader is stopped, causing the slave's counts significantly higher. To temporarily solve this,
274  # let's set the tolerance rate to 80%.
275  # TODO: Revert the code for tolerance once the throttling mechanism is fixed.
276  index=0
277  valid_counts=0
278  invalid_counts=0
279  tolerance_rate=0.8
280  while IFS= read -r line
281  do
282    cycles=$(echo $line | awk '{for(i=1;i<=NF;i++) if($i=="cycles:") print $(i-1)}')
283    if [ $(($index%2)) -ne 0 ] && [ ${cycles}x != ${prev_cycles}x ]
284    then
285      invalid_counts=$(($invalid_counts+1))
286    else
287      valid_counts=$(($valid_counts+1))
288    fi
289    index=$(($index+1))
290    prev_cycles=$cycles
291  done < "${script_output}"
292  total_counts=$(bc <<< "$invalid_counts+$valid_counts")
293  if (( $(bc <<< "$total_counts <= 0") ))
294  then
295    echo "Leader sampling [No sample generated]"
296    err=1
297    return
298  fi
299  isok=$(bc <<< "scale=2; if (($invalid_counts/$total_counts) < (1-$tolerance_rate)) { 0 } else { 1 };")
300  if [ $isok -eq 1 ]
301  then
302     echo "Leader sampling [Failed inconsistent cycles count]"
303     err=1
304  else
305    echo "Basic leader sampling test [Success]"
306  fi
307}
308
309test_topdown_leader_sampling() {
310  echo "Topdown leader sampling test"
311  if ! perf stat -e "{slots,topdown-retiring}" true 2> /dev/null
312  then
313    echo "Topdown leader sampling [Skipped event parsing failed]"
314    return
315  fi
316  if ! perf record -o "${perfdata}" -e "{instructions,slots,topdown-retiring}:S" true 2> /dev/null
317  then
318    echo "Topdown leader sampling [Failed topdown events not reordered correctly]"
319    err=1
320    return
321  fi
322  echo "Topdown leader sampling test [Success]"
323}
324
325test_precise_max() {
326  local -i skipped=0
327
328  echo "precise_max attribute test"
329  # Just to make sure event cycles is supported for sampling
330  if perf record -o "${perfdata}" -e "cycles" true 2> /dev/null
331  then
332    if ! perf record -o "${perfdata}" -e "cycles:P" true 2> /dev/null
333    then
334      echo "precise_max attribute [Failed cycles:P event]"
335      err=1
336      return
337    fi
338  else
339    echo "precise_max attribute [Skipped no cycles:P event]"
340    ((skipped+=1))
341  fi
342  # On s390 event instructions is not supported for perf record
343  if perf record -o "${perfdata}" -e "instructions" true 2> /dev/null
344  then
345    # On AMD, cycles and instructions events are treated differently
346    if ! perf record -o "${perfdata}" -e "instructions:P" true 2> /dev/null
347    then
348      echo "precise_max attribute [Failed instructions:P event]"
349      err=1
350      return
351    fi
352  else
353    echo "precise_max attribute [Skipped no instructions:P event]"
354    ((skipped+=1))
355  fi
356  if [ $skipped -eq 2 ]
357  then
358    echo "precise_max attribute [Skipped no hardware events]"
359  else
360    echo "precise_max attribute test [Success]"
361  fi
362}
363
364test_callgraph() {
365  echo "Callgraph test"
366
367  case $(uname -m)
368  in s390x)
369       cmd_flags="--call-graph dwarf -e cpu-clock";;
370     *)
371       cmd_flags="-g";;
372  esac
373
374  if ! perf record -o "${perfdata}" $cmd_flags perf test -w brstack
375  then
376    echo "Callgraph test [Failed missing output]"
377    err=1
378    return
379  fi
380
381  if ! perf report -i "${perfdata}" 2>&1 | grep "${testsym2}"
382  then
383    echo "Callgraph test [Failed missing symbol]"
384    err=1
385    return
386  fi
387
388  echo "Callgraph test [Success]"
389}
390
391# raise the limit of file descriptors to minimum
392if [[ $default_fd_limit -lt $min_fd_limit ]]; then
393       ulimit -Sn $min_fd_limit
394fi
395
396test_per_thread
397test_register_capture
398test_system_wide
399test_workload
400test_branch_counter
401test_cgroup
402test_uid
403test_leader_sampling
404test_topdown_leader_sampling
405test_precise_max
406test_callgraph
407
408# restore the default value
409ulimit -Sn $default_fd_limit
410
411cleanup
412exit $err
413