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

[Arnaldo Carvalho de Melo]

On Wed, Jul 24, 2024 at 01:20:27PM -0700, Namhyung Kim wrote:
> On Wed, Jul 24, 2024 at 04:32:16PM -0300, Arnaldo Carvalho de Melo wrote:
> > 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?
> 
> There's a 'perf record sample filtering (by BPF) tests'.
> 
>   $ ./perf test -vv filtering
>    95: perf record sample filtering (by BPF) tests:
>   --- start ---
>   test child forked, pid 1042594
>   Checking BPF-filter privilege
>   try 'sudo perf record --setup-filter pin' first.
>   bpf-filter test [Skipped permission]
>   ---- end(-2) ----
>    95: perf record sample filtering (by BPF) tests                     : Skip
> 
> > 
> > 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:~#
> 
> Do you say it's failing after the first patch?  It looks like the atom

yes

> CPU doesn't support PERF_SAMPLE_WEIGHT and should fail already.

I tried with 'cycles:p', 'cpu_atom/cycles/p' and with
'cpu_core/cycles/p', with and without -W (to use the warning advice)
will try again tomorrow.

- Arnaldo
 
> The above test doesn't check the weight field FYI.
> 
> > 
> > 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?
> 
> I don't know.. is it changed with this patch?
> 
> Thanks,
> Namhyung