Perf-stat tool does not support ipc and ipc_rate monitoring on NVIDIA Grace system
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
linux-nvidia-6.5 (Ubuntu) |
New
|
Undecided
|
Unassigned |
Bug Description
PROBLEM:
While trying to monitor the ipc group for using perf-stat tool, despite it being supported, I get the following errors:
$ sudo perf list | grep ipc
ipc
ipc_rate
retired_ipc
spec_ipc
$ sudo -S perf stat -a -M ipc -- sudo -S stress-ng --cpu 0 -t 10s
Error:
The sys_perf_
/bin/dmesg | grep -i perf may provide additional information.
$ sudo -S perf stat -a -M ipc_rate -- sudo -S stress-ng --cpu 0 -t 10s
Error:
The sys_perf_
/bin/dmesg | grep -i perf may provide additional information.
However, I can get the groups retired_ipc and spec_ipc to work:
$ sudo -S perf stat -a -M retired_ipc -- sudo -S stress-ng --cpu 0 -t 10s
Value 0 contains non-numeric: ' '
Performance counter stats for 'system wide':
96,818,964 INST_RETIRED # 0.58 retired_ipc
166,601,455 CPU_CYCLES
0.013516186 seconds time elapsed
$ sudo -S perf stat -a -M spec_ipc -- sudo -S stress-ng --cpu 0 -t 10s
Value 0 contains non-numeric: ' '
Performance counter stats for 'system wide':
91,053,297 INST_SPEC # 0.58 spec_ipc
156,558,810 CPU_CYCLES
0.009877355 seconds time elapsed
SOLUTION:
Please accept the pull request which cherry-picks the following two upstream commits:
d43f54912101971
for N2 r0p3
4473949074c3507
events using Arm telemetry repo