Re: [PATCH v3 1/8] perf bpf-filter: Make filters map a single entry hashmap

[Arnaldo Carvalho de Melo]

On Wed, Jul 03, 2024 at 03:30:28PM -0700, Namhyung Kim wrote:
> And the value is now an array.  This is to support multiple filter
> entries in the map later.
> 
> No functional changes intended.

Hey how can we test this feature these days?

With this first patch applied:

root@number:~# perf record -a -W -e cycles:p --filter 'period > 100 || weight > 0' sleep 1
Error: cpu_atom/cycles/p event does not have PERF_SAMPLE_WEIGHT
 Hint: please add -W option to perf record
failed to set filter "BPF" on event cpu_atom/cycles/p with 95 (Operation not supported)
root@number:~# perf record -a -W -e cpu_core/cycles/p --filter 'period > 100 || weight > 0' sleep 1
Error: cpu_core/cycles/p event does not have PERF_SAMPLE_WEIGHT
 Hint: please add -W option to perf record
failed to set filter "BPF" on event cpu_core/cycles/p with 95 (Operation not supported)
root@number:~# perf record -a -W -e cpu_atom/cycles/p --filter 'period > 100 || weight > 0' sleep 1
Error: cpu_atom/cycles/p event does not have PERF_SAMPLE_WEIGHT
 Hint: please add -W option to perf record
failed to set filter "BPF" on event cpu_atom/cycles/p with 95 (Operation not supported)
root@number:~#

Interesting, it is taking a long time on the BPF prog load:

bpf(BPF_MAP_UPDATE_ELEM, {map_fd=49, key=0x7ffcc85a545c, value=0x7fee34bc2000, flags=BPF_ANY}, 32) = 0
bpf(BPF_PROG_LOAD, {prog_type=BPF_PROG_TYPE_PERF_EVENT, insn_cnt=335, insns=0xd1e6480, license="Dual BSD/GPL", log_level=0, log_size=0, log_buf=NULL, kern_version=KERNEL_VERSION(6, 9, 9), prog_flags=0, prog_name="perf_sample_fil", prog_ifindex=0, expected_attach_type=BPF_CGROUP_INET_INGRESS, prog_btf_fd=50, func_info_rec_size=8, func_info=0xd1b9c80, func_info_cnt=1, line_info_rec_size=16, line_info=0xd1e5300, line_info_cnt=135, attach_btf_id=0, attach_prog_fd=0, fd_array=NULL}, 148^Cstrace: Process 2110180 detached
 <detached ...>

<HERE it takes an unusual time, even returning after I cancelled the strace session>

root@number:~# 
root@number:~# Error: cpu_atom/cycles/p event does not have PERF_SAMPLE_WEIGHT
 Hint: please add -W option to perf record
failed to set filter "BPF" on event cpu_atom/cycles/p with 11 (Resource temporarily unavailable)

root@number:~#


root@number:~# uname -a
Linux number 6.9.9-100.fc39.x86_64 #1 SMP PREEMPT_DYNAMIC Thu Jul 11 19:26:10 UTC 2024 x86_64 GNU/Linux
root@number:~#

root@number:~# perf -v
perf version 6.10.g5510fb5c79e9
root@number:~# 

⬢[acme@toolbox perf-tools-next]$ git log --oneline -10
5510fb5c79e9f500 (HEAD -> perf-tools-next) perf annotate: Set instruction name to be used with insn-stat when using raw instruction
b35a86e53eb496ea perf annotate: Add support to use libcapstone in powerpc
f2dc60d11290d53e perf annotate: Use capstone_init and remove open_capstone_handle from disasm.c
c5bcba602eeee554 perf annotate: Make capstone_init non-static so that it can be used during symbol disassemble
eef369c562510092 perf annotate: Update instruction tracking for powerpc
282701f1d77a3bdb perf annotate: Add more instructions for instruction tracking
758ee468ce5721e4 perf annotate: Add some of the arithmetic instructions to support instruction tracking in powerpc
e8e7c1b6a9572bab perf annotate: Add support to identify memory instructions of opcode 31 in powerpc
3b3a0f04c1c6cd10 perf annotate: Add parse function for memory instructions in powerpc
a159d2acd44e707f perf annotate: Update parameters for reg extract functions to use raw instruction on powerpc
⬢[acme@toolbox perf-tools-next]$ 

Ideas?

- Arnaldo