mbox series

[RFC/PATCHSET,0/7] perf record: Implement BPF sample filter (v1)

Message ID 20230214050452.26390-1-namhyung@kernel.org (mailing list archive)
Headers show
Series perf record: Implement BPF sample filter (v1) | expand

Message

Namhyung Kim Feb. 14, 2023, 5:04 a.m. UTC
Hello,

There have been requests for more sophisticated perf event sample
filtering based on the sample data.  Recently the kernel added BPF
programs can access perf sample data and this is the userspace part
to enable such a filtering.

This still has some rough edges and needs more improvements.  But
I'd like to share the current work and get some feedback for the
directions and idea for further improvements.

The kernel changes are in the tip.git tree (perf/core branch) for now.
perf record has --filter option to set filters on the last specified
event in the command line.  It worked only for tracepoints and Intel
PT events so far.  This patchset extends it to have 'bpf:' prefix in
order to enable the general sample filters using BPF for any events.

A new filter expression parser was added (using flex/bison) to process
the filter string.  Right now, it only accepts very simple expressions
separated by comma.  I'd like to keep the filter expression as simple
as possible.

It requires samples satisfy all the filter expressions otherwise it'd
drop the sample.  IOW filter expressions are connected with logical AND
operations implicitly.

Essentially the BPF filter expression is:

  "bpf:" <term> <operator> <value> ("," <term> <operator> <value>)*

The <term> can be one of:
  ip, id, tid, pid, cpu, time, addr, period, txn, weight, phys_addr,
  code_pgsz, data_pgsz, weight1, weight2, weight3, ins_lat, retire_lat,
  p_stage_cyc, mem_op, mem_lvl, mem_snoop, mem_remote, mem_lock,
  mem_dtlb, mem_blk, mem_hops

The <operator> can be one of:
  ==, !=, >, >=, <, <=, &

The <value> can be one of:
  <number> (for any term)
  na, load, store, pfetch, exec (for mem_op)
  l1, l2, l3, l4, cxl, io, any_cache, lfb, ram, pmem (for mem_lvl)
  na, none, hit, miss, hitm, fwd, peer (for mem_snoop)
  remote (for mem_remote)
  na, locked (for mem_locked)
  na, l1_hit, l1_miss, l2_hit, l2_miss, any_hit, any_miss, walk, fault (for mem_dtlb)
  na, by_data, by_addr (for mem_blk)
  hops0, hops1, hops2, hops3 (for mem_hops)

I plan to improve it with range expressions like for ip or addr and it
should support symbols like the existing addr-filters.  Also cgroup
should understand and convert cgroup names to IDs.

Let's take a look at some examples.  The following is to profile a user
program on the command line.  When the frequency mode is used, it starts
with a very small period (i.e. 1) and adjust it on every interrupt (NMI)
to catch up the given frequency.

  $ ./perf record -- ./perf test -w noploop
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.263 MB perf.data (4006 samples) ]

  $ ./perf script -F pid,period,event,ip,sym | head
  36695          1 cycles:  ffffffffbab12ddd perf_event_exec
  36695          1 cycles:  ffffffffbab12ddd perf_event_exec
  36695          5 cycles:  ffffffffbab12ddd perf_event_exec
  36695         46 cycles:  ffffffffbab12de5 perf_event_exec
  36695       1163 cycles:  ffffffffba80a0eb x86_pmu_disable_all
  36695       1304 cycles:  ffffffffbaa19507 __hrtimer_get_next_event
  36695       8143 cycles:  ffffffffbaa186f9 __run_timers
  36695      69040 cycles:  ffffffffbaa0c393 rcu_segcblist_ready_cbs
  36695     355117 cycles:            4b0da4 noploop
  36695     321861 cycles:            4b0da4 noploop

If you want to skip the first few samples that have small periods, you
can do like this (note it requires root due to BPF).

  $ sudo ./perf record -e cycles --filter 'bpf: period > 10000' -- ./perf test -w noploop
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.262 MB perf.data (3990 samples) ]

  $ sudo ./perf script -F pid,period,event,ip,sym | head
  39524      58253 cycles:  ffffffffba97dac0 update_rq_clock
  39524     232657 cycles:            4b0da2 noploop
  39524     210981 cycles:            4b0da2 noploop
  39524     282882 cycles:            4b0da4 noploop
  39524     392180 cycles:            4b0da4 noploop
  39524     456058 cycles:            4b0da4 noploop
  39524     415196 cycles:            4b0da2 noploop
  39524     462721 cycles:            4b0da4 noploop
  39524     526272 cycles:            4b0da2 noploop
  39524     565569 cycles:            4b0da4 noploop

Maybe more useful example is when it deals with precise memory events.
On AMD processors with IBS, you can filter only memory load with L1
dTLB is missed like below.

  $ sudo ./perf record -ad -e ibs_op//p \
  > --filter 'bpf: mem_op == load, mem_dtlb > l1_hit' sleep 1
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 1.338 MB perf.data (15 samples) ]

  $ sudo ./perf script -F data_src | head
          51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
          49080142 |OP LOAD|LVL L1 hit|SNP N/A|TLB L2 hit|LCK N/A|BLK  N/A
          51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
          51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
          51088842 |OP LOAD|LVL L3 or Remote Cache (1 hop) hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
          51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
          51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
          51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
          49080442 |OP LOAD|LVL L2 hit|SNP N/A|TLB L2 hit|LCK N/A|BLK  N/A
          51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A

You can also check the number of dropped samples in LOST_SAMPLES events
using perf report --stat command.

  $ sudo ./perf report --stat

  Aggregated stats:
             TOTAL events:      16066
              MMAP events:         22  ( 0.1%)
              COMM events:       4166  (25.9%)
              EXIT events:          1  ( 0.0%)
          THROTTLE events:        816  ( 5.1%)
        UNTHROTTLE events:        613  ( 3.8%)
              FORK events:       4165  (25.9%)
            SAMPLE events:         15  ( 0.1%)
             MMAP2 events:       6133  (38.2%)
      LOST_SAMPLES events:          1  ( 0.0%)
           KSYMBOL events:         69  ( 0.4%)
         BPF_EVENT events:         57  ( 0.4%)
    FINISHED_ROUND events:          3  ( 0.0%)
          ID_INDEX events:          1  ( 0.0%)
        THREAD_MAP events:          1  ( 0.0%)
           CPU_MAP events:          1  ( 0.0%)
         TIME_CONV events:          1  ( 0.0%)
     FINISHED_INIT events:          1  ( 0.0%)
  ibs_op//p stats:
            SAMPLE events:         15
      LOST_SAMPLES events:       3991

Note that the total aggregated stats show 1 LOST_SAMPLES event but
per event stats show 3991 events because it's the actual number of
dropped samples while the aggregated stats has the number of record.
Maybe we need to change the per-event stats to 'LOST_SAMPLES count'
to avoid the confusion.

The code is available at 'perf/bpf-filter-v1' branch in my tree.

  git://git.kernel.org/pub/scm/linux/kernel/git/namhyung/linux-perf.git

Again, you need tip/perf/core kernel for this to work.
Any feedback is welcome.

Thanks,
Namhyung

Namhyung Kim (7):
  perf bpf filter: Introduce basic BPF filter expression
  perf bpf filter: Implement event sample filtering
  perf record: Add BPF event filter support
  perf record: Record dropped sample count
  perf bpf filter: Add 'pid' sample data support
  perf bpf filter: Add more weight sample data support
  perf bpf filter: Add data_src sample data support

 tools/perf/Documentation/perf-record.txt     |  10 +-
 tools/perf/Makefile.perf                     |   2 +-
 tools/perf/builtin-record.c                  |  46 ++++--
 tools/perf/util/Build                        |  16 ++
 tools/perf/util/bpf-filter.c                 | 117 ++++++++++++++
 tools/perf/util/bpf-filter.h                 |  48 ++++++
 tools/perf/util/bpf-filter.l                 | 146 ++++++++++++++++++
 tools/perf/util/bpf-filter.y                 |  55 +++++++
 tools/perf/util/bpf_counter.c                |   3 +-
 tools/perf/util/bpf_skel/sample-filter.h     |  25 +++
 tools/perf/util/bpf_skel/sample_filter.bpf.c | 152 +++++++++++++++++++
 tools/perf/util/evsel.c                      |   2 +
 tools/perf/util/evsel.h                      |   7 +-
 tools/perf/util/parse-events.c               |   4 +
 tools/perf/util/session.c                    |   3 +-
 15 files changed, 615 insertions(+), 21 deletions(-)
 create mode 100644 tools/perf/util/bpf-filter.c
 create mode 100644 tools/perf/util/bpf-filter.h
 create mode 100644 tools/perf/util/bpf-filter.l
 create mode 100644 tools/perf/util/bpf-filter.y
 create mode 100644 tools/perf/util/bpf_skel/sample-filter.h
 create mode 100644 tools/perf/util/bpf_skel/sample_filter.bpf.c


base-commit: 37f322cd58d81a9d46456531281c908de9ef6e42

Comments

Ian Rogers Feb. 14, 2023, 4:57 p.m. UTC | #1
On Mon, Feb 13, 2023 at 9:05 PM Namhyung Kim <namhyung@kernel.org> wrote:
>
> Hello,
>
> There have been requests for more sophisticated perf event sample
> filtering based on the sample data.  Recently the kernel added BPF
> programs can access perf sample data and this is the userspace part
> to enable such a filtering.
>
> This still has some rough edges and needs more improvements.  But
> I'd like to share the current work and get some feedback for the
> directions and idea for further improvements.
>
> The kernel changes are in the tip.git tree (perf/core branch) for now.
> perf record has --filter option to set filters on the last specified
> event in the command line.  It worked only for tracepoints and Intel
> PT events so far.  This patchset extends it to have 'bpf:' prefix in
> order to enable the general sample filters using BPF for any events.
>
> A new filter expression parser was added (using flex/bison) to process
> the filter string.  Right now, it only accepts very simple expressions
> separated by comma.  I'd like to keep the filter expression as simple
> as possible.
>
> It requires samples satisfy all the filter expressions otherwise it'd
> drop the sample.  IOW filter expressions are connected with logical AND
> operations implicitly.
>
> Essentially the BPF filter expression is:
>
>   "bpf:" <term> <operator> <value> ("," <term> <operator> <value>)*
>
> The <term> can be one of:
>   ip, id, tid, pid, cpu, time, addr, period, txn, weight, phys_addr,
>   code_pgsz, data_pgsz, weight1, weight2, weight3, ins_lat, retire_lat,
>   p_stage_cyc, mem_op, mem_lvl, mem_snoop, mem_remote, mem_lock,
>   mem_dtlb, mem_blk, mem_hops
>
> The <operator> can be one of:
>   ==, !=, >, >=, <, <=, &
>
> The <value> can be one of:
>   <number> (for any term)
>   na, load, store, pfetch, exec (for mem_op)
>   l1, l2, l3, l4, cxl, io, any_cache, lfb, ram, pmem (for mem_lvl)
>   na, none, hit, miss, hitm, fwd, peer (for mem_snoop)
>   remote (for mem_remote)
>   na, locked (for mem_locked)
>   na, l1_hit, l1_miss, l2_hit, l2_miss, any_hit, any_miss, walk, fault (for mem_dtlb)
>   na, by_data, by_addr (for mem_blk)
>   hops0, hops1, hops2, hops3 (for mem_hops)
>
> I plan to improve it with range expressions like for ip or addr and it
> should support symbols like the existing addr-filters.  Also cgroup
> should understand and convert cgroup names to IDs.
>
> Let's take a look at some examples.  The following is to profile a user
> program on the command line.  When the frequency mode is used, it starts
> with a very small period (i.e. 1) and adjust it on every interrupt (NMI)
> to catch up the given frequency.
>
>   $ ./perf record -- ./perf test -w noploop
>   [ perf record: Woken up 1 times to write data ]
>   [ perf record: Captured and wrote 0.263 MB perf.data (4006 samples) ]
>
>   $ ./perf script -F pid,period,event,ip,sym | head
>   36695          1 cycles:  ffffffffbab12ddd perf_event_exec
>   36695          1 cycles:  ffffffffbab12ddd perf_event_exec
>   36695          5 cycles:  ffffffffbab12ddd perf_event_exec
>   36695         46 cycles:  ffffffffbab12de5 perf_event_exec
>   36695       1163 cycles:  ffffffffba80a0eb x86_pmu_disable_all
>   36695       1304 cycles:  ffffffffbaa19507 __hrtimer_get_next_event
>   36695       8143 cycles:  ffffffffbaa186f9 __run_timers
>   36695      69040 cycles:  ffffffffbaa0c393 rcu_segcblist_ready_cbs
>   36695     355117 cycles:            4b0da4 noploop
>   36695     321861 cycles:            4b0da4 noploop
>
> If you want to skip the first few samples that have small periods, you
> can do like this (note it requires root due to BPF).
>
>   $ sudo ./perf record -e cycles --filter 'bpf: period > 10000' -- ./perf test -w noploop
>   [ perf record: Woken up 1 times to write data ]
>   [ perf record: Captured and wrote 0.262 MB perf.data (3990 samples) ]
>
>   $ sudo ./perf script -F pid,period,event,ip,sym | head
>   39524      58253 cycles:  ffffffffba97dac0 update_rq_clock
>   39524     232657 cycles:            4b0da2 noploop
>   39524     210981 cycles:            4b0da2 noploop
>   39524     282882 cycles:            4b0da4 noploop
>   39524     392180 cycles:            4b0da4 noploop
>   39524     456058 cycles:            4b0da4 noploop
>   39524     415196 cycles:            4b0da2 noploop
>   39524     462721 cycles:            4b0da4 noploop
>   39524     526272 cycles:            4b0da2 noploop
>   39524     565569 cycles:            4b0da4 noploop
>
> Maybe more useful example is when it deals with precise memory events.
> On AMD processors with IBS, you can filter only memory load with L1
> dTLB is missed like below.
>
>   $ sudo ./perf record -ad -e ibs_op//p \
>   > --filter 'bpf: mem_op == load, mem_dtlb > l1_hit' sleep 1
>   [ perf record: Woken up 1 times to write data ]
>   [ perf record: Captured and wrote 1.338 MB perf.data (15 samples) ]
>
>   $ sudo ./perf script -F data_src | head
>           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
>           49080142 |OP LOAD|LVL L1 hit|SNP N/A|TLB L2 hit|LCK N/A|BLK  N/A
>           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
>           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
>           51088842 |OP LOAD|LVL L3 or Remote Cache (1 hop) hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
>           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
>           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
>           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
>           49080442 |OP LOAD|LVL L2 hit|SNP N/A|TLB L2 hit|LCK N/A|BLK  N/A
>           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
>
> You can also check the number of dropped samples in LOST_SAMPLES events
> using perf report --stat command.
>
>   $ sudo ./perf report --stat
>
>   Aggregated stats:
>              TOTAL events:      16066
>               MMAP events:         22  ( 0.1%)
>               COMM events:       4166  (25.9%)
>               EXIT events:          1  ( 0.0%)
>           THROTTLE events:        816  ( 5.1%)
>         UNTHROTTLE events:        613  ( 3.8%)
>               FORK events:       4165  (25.9%)
>             SAMPLE events:         15  ( 0.1%)
>              MMAP2 events:       6133  (38.2%)
>       LOST_SAMPLES events:          1  ( 0.0%)
>            KSYMBOL events:         69  ( 0.4%)
>          BPF_EVENT events:         57  ( 0.4%)
>     FINISHED_ROUND events:          3  ( 0.0%)
>           ID_INDEX events:          1  ( 0.0%)
>         THREAD_MAP events:          1  ( 0.0%)
>            CPU_MAP events:          1  ( 0.0%)
>          TIME_CONV events:          1  ( 0.0%)
>      FINISHED_INIT events:          1  ( 0.0%)
>   ibs_op//p stats:
>             SAMPLE events:         15
>       LOST_SAMPLES events:       3991
>
> Note that the total aggregated stats show 1 LOST_SAMPLES event but
> per event stats show 3991 events because it's the actual number of
> dropped samples while the aggregated stats has the number of record.
> Maybe we need to change the per-event stats to 'LOST_SAMPLES count'
> to avoid the confusion.
>
> The code is available at 'perf/bpf-filter-v1' branch in my tree.
>
>   git://git.kernel.org/pub/scm/linux/kernel/git/namhyung/linux-perf.git
>
> Again, you need tip/perf/core kernel for this to work.
> Any feedback is welcome.

This is great! I wonder about related clean up:

 - can we remove BPF events as this is a better feature?
   - I believe BPF events are flaky, seldom used (with the exception
of the augmented syscalls for perf trace, which really should move to
a BPF skeleton as most people don't know how to use it) and they add a
bunch of complexity. A particular complexity I care about is that the
path separator forward slash ('/') is also the modifier separator for
events.

 - what will happen with multiple events/metrics? Perhaps there should
be a way of listing filters so that each filter applies to the
appropriate event in the event list, like cgroups and -G. For metrics
we shuffle the list of events and so maybe the filters need some way
to specify which event they apply to.

 - It feels like there should be some BPF way of overcoming the fixed
length number of filters so it is still bounded but not a hardcoded
number.

Thanks,
Ian


> Thanks,
> Namhyung
>
> Namhyung Kim (7):
>   perf bpf filter: Introduce basic BPF filter expression
>   perf bpf filter: Implement event sample filtering
>   perf record: Add BPF event filter support
>   perf record: Record dropped sample count
>   perf bpf filter: Add 'pid' sample data support
>   perf bpf filter: Add more weight sample data support
>   perf bpf filter: Add data_src sample data support
>
>  tools/perf/Documentation/perf-record.txt     |  10 +-
>  tools/perf/Makefile.perf                     |   2 +-
>  tools/perf/builtin-record.c                  |  46 ++++--
>  tools/perf/util/Build                        |  16 ++
>  tools/perf/util/bpf-filter.c                 | 117 ++++++++++++++
>  tools/perf/util/bpf-filter.h                 |  48 ++++++
>  tools/perf/util/bpf-filter.l                 | 146 ++++++++++++++++++
>  tools/perf/util/bpf-filter.y                 |  55 +++++++
>  tools/perf/util/bpf_counter.c                |   3 +-
>  tools/perf/util/bpf_skel/sample-filter.h     |  25 +++
>  tools/perf/util/bpf_skel/sample_filter.bpf.c | 152 +++++++++++++++++++
>  tools/perf/util/evsel.c                      |   2 +
>  tools/perf/util/evsel.h                      |   7 +-
>  tools/perf/util/parse-events.c               |   4 +
>  tools/perf/util/session.c                    |   3 +-
>  15 files changed, 615 insertions(+), 21 deletions(-)
>  create mode 100644 tools/perf/util/bpf-filter.c
>  create mode 100644 tools/perf/util/bpf-filter.h
>  create mode 100644 tools/perf/util/bpf-filter.l
>  create mode 100644 tools/perf/util/bpf-filter.y
>  create mode 100644 tools/perf/util/bpf_skel/sample-filter.h
>  create mode 100644 tools/perf/util/bpf_skel/sample_filter.bpf.c
>
>
> base-commit: 37f322cd58d81a9d46456531281c908de9ef6e42
> --
> 2.39.1.581.gbfd45094c4-goog
>
Namhyung Kim Feb. 14, 2023, 6:01 p.m. UTC | #2
Hi Ian,

On Tue, Feb 14, 2023 at 8:58 AM Ian Rogers <irogers@google.com> wrote:
>
> On Mon, Feb 13, 2023 at 9:05 PM Namhyung Kim <namhyung@kernel.org> wrote:
> >
> > Hello,
> >
> > There have been requests for more sophisticated perf event sample
> > filtering based on the sample data.  Recently the kernel added BPF
> > programs can access perf sample data and this is the userspace part
> > to enable such a filtering.
> >
> > This still has some rough edges and needs more improvements.  But
> > I'd like to share the current work and get some feedback for the
> > directions and idea for further improvements.
> >
> > The kernel changes are in the tip.git tree (perf/core branch) for now.
> > perf record has --filter option to set filters on the last specified
> > event in the command line.  It worked only for tracepoints and Intel
> > PT events so far.  This patchset extends it to have 'bpf:' prefix in
> > order to enable the general sample filters using BPF for any events.
> >
> > A new filter expression parser was added (using flex/bison) to process
> > the filter string.  Right now, it only accepts very simple expressions
> > separated by comma.  I'd like to keep the filter expression as simple
> > as possible.
> >
> > It requires samples satisfy all the filter expressions otherwise it'd
> > drop the sample.  IOW filter expressions are connected with logical AND
> > operations implicitly.
> >
> > Essentially the BPF filter expression is:
> >
> >   "bpf:" <term> <operator> <value> ("," <term> <operator> <value>)*
> >
> > The <term> can be one of:
> >   ip, id, tid, pid, cpu, time, addr, period, txn, weight, phys_addr,
> >   code_pgsz, data_pgsz, weight1, weight2, weight3, ins_lat, retire_lat,
> >   p_stage_cyc, mem_op, mem_lvl, mem_snoop, mem_remote, mem_lock,
> >   mem_dtlb, mem_blk, mem_hops
> >
> > The <operator> can be one of:
> >   ==, !=, >, >=, <, <=, &
> >
> > The <value> can be one of:
> >   <number> (for any term)
> >   na, load, store, pfetch, exec (for mem_op)
> >   l1, l2, l3, l4, cxl, io, any_cache, lfb, ram, pmem (for mem_lvl)
> >   na, none, hit, miss, hitm, fwd, peer (for mem_snoop)
> >   remote (for mem_remote)
> >   na, locked (for mem_locked)
> >   na, l1_hit, l1_miss, l2_hit, l2_miss, any_hit, any_miss, walk, fault (for mem_dtlb)
> >   na, by_data, by_addr (for mem_blk)
> >   hops0, hops1, hops2, hops3 (for mem_hops)
> >
> > I plan to improve it with range expressions like for ip or addr and it
> > should support symbols like the existing addr-filters.  Also cgroup
> > should understand and convert cgroup names to IDs.
> >
> > Let's take a look at some examples.  The following is to profile a user
> > program on the command line.  When the frequency mode is used, it starts
> > with a very small period (i.e. 1) and adjust it on every interrupt (NMI)
> > to catch up the given frequency.
> >
> >   $ ./perf record -- ./perf test -w noploop
> >   [ perf record: Woken up 1 times to write data ]
> >   [ perf record: Captured and wrote 0.263 MB perf.data (4006 samples) ]
> >
> >   $ ./perf script -F pid,period,event,ip,sym | head
> >   36695          1 cycles:  ffffffffbab12ddd perf_event_exec
> >   36695          1 cycles:  ffffffffbab12ddd perf_event_exec
> >   36695          5 cycles:  ffffffffbab12ddd perf_event_exec
> >   36695         46 cycles:  ffffffffbab12de5 perf_event_exec
> >   36695       1163 cycles:  ffffffffba80a0eb x86_pmu_disable_all
> >   36695       1304 cycles:  ffffffffbaa19507 __hrtimer_get_next_event
> >   36695       8143 cycles:  ffffffffbaa186f9 __run_timers
> >   36695      69040 cycles:  ffffffffbaa0c393 rcu_segcblist_ready_cbs
> >   36695     355117 cycles:            4b0da4 noploop
> >   36695     321861 cycles:            4b0da4 noploop
> >
> > If you want to skip the first few samples that have small periods, you
> > can do like this (note it requires root due to BPF).
> >
> >   $ sudo ./perf record -e cycles --filter 'bpf: period > 10000' -- ./perf test -w noploop
> >   [ perf record: Woken up 1 times to write data ]
> >   [ perf record: Captured and wrote 0.262 MB perf.data (3990 samples) ]
> >
> >   $ sudo ./perf script -F pid,period,event,ip,sym | head
> >   39524      58253 cycles:  ffffffffba97dac0 update_rq_clock
> >   39524     232657 cycles:            4b0da2 noploop
> >   39524     210981 cycles:            4b0da2 noploop
> >   39524     282882 cycles:            4b0da4 noploop
> >   39524     392180 cycles:            4b0da4 noploop
> >   39524     456058 cycles:            4b0da4 noploop
> >   39524     415196 cycles:            4b0da2 noploop
> >   39524     462721 cycles:            4b0da4 noploop
> >   39524     526272 cycles:            4b0da2 noploop
> >   39524     565569 cycles:            4b0da4 noploop
> >
> > Maybe more useful example is when it deals with precise memory events.
> > On AMD processors with IBS, you can filter only memory load with L1
> > dTLB is missed like below.
> >
> >   $ sudo ./perf record -ad -e ibs_op//p \
> >   > --filter 'bpf: mem_op == load, mem_dtlb > l1_hit' sleep 1
> >   [ perf record: Woken up 1 times to write data ]
> >   [ perf record: Captured and wrote 1.338 MB perf.data (15 samples) ]
> >
> >   $ sudo ./perf script -F data_src | head
> >           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> >           49080142 |OP LOAD|LVL L1 hit|SNP N/A|TLB L2 hit|LCK N/A|BLK  N/A
> >           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> >           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> >           51088842 |OP LOAD|LVL L3 or Remote Cache (1 hop) hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> >           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> >           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> >           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> >           49080442 |OP LOAD|LVL L2 hit|SNP N/A|TLB L2 hit|LCK N/A|BLK  N/A
> >           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> >
> > You can also check the number of dropped samples in LOST_SAMPLES events
> > using perf report --stat command.
> >
> >   $ sudo ./perf report --stat
> >
> >   Aggregated stats:
> >              TOTAL events:      16066
> >               MMAP events:         22  ( 0.1%)
> >               COMM events:       4166  (25.9%)
> >               EXIT events:          1  ( 0.0%)
> >           THROTTLE events:        816  ( 5.1%)
> >         UNTHROTTLE events:        613  ( 3.8%)
> >               FORK events:       4165  (25.9%)
> >             SAMPLE events:         15  ( 0.1%)
> >              MMAP2 events:       6133  (38.2%)
> >       LOST_SAMPLES events:          1  ( 0.0%)
> >            KSYMBOL events:         69  ( 0.4%)
> >          BPF_EVENT events:         57  ( 0.4%)
> >     FINISHED_ROUND events:          3  ( 0.0%)
> >           ID_INDEX events:          1  ( 0.0%)
> >         THREAD_MAP events:          1  ( 0.0%)
> >            CPU_MAP events:          1  ( 0.0%)
> >          TIME_CONV events:          1  ( 0.0%)
> >      FINISHED_INIT events:          1  ( 0.0%)
> >   ibs_op//p stats:
> >             SAMPLE events:         15
> >       LOST_SAMPLES events:       3991
> >
> > Note that the total aggregated stats show 1 LOST_SAMPLES event but
> > per event stats show 3991 events because it's the actual number of
> > dropped samples while the aggregated stats has the number of record.
> > Maybe we need to change the per-event stats to 'LOST_SAMPLES count'
> > to avoid the confusion.
> >
> > The code is available at 'perf/bpf-filter-v1' branch in my tree.
> >
> >   git://git.kernel.org/pub/scm/linux/kernel/git/namhyung/linux-perf.git
> >
> > Again, you need tip/perf/core kernel for this to work.
> > Any feedback is welcome.
>
> This is great! I wonder about related clean up:
>
>  - can we remove BPF events as this is a better feature?
>    - I believe BPF events are flaky, seldom used (with the exception
> of the augmented syscalls for perf trace, which really should move to
> a BPF skeleton as most people don't know how to use it) and they add a
> bunch of complexity. A particular complexity I care about is that the
> path separator forward slash ('/') is also the modifier separator for
> events.

Well.. I actually never tried the BPF events myself :)
I think we can deprecate it and get rid of it once the perf trace
conversion is done.

>
>  - what will happen with multiple events/metrics? Perhaps there should
> be a way of listing filters so that each filter applies to the
> appropriate event in the event list, like cgroups and -G. For metrics
> we shuffle the list of events and so maybe the filters need some way
> to specify which event they apply to.

For now, it's applied to the last event specified by '-e' before the fitter.
As it's local to the event, you should be able to use appropriate one
for each event.  I didn't think about the metrics as it's for perf record
only.

>
>  - It feels like there should be some BPF way of overcoming the fixed
> length number of filters so it is still bounded but not a hardcoded
> number.

Maybe.. but note that the hardcoded max is just for the verifier.
At runtime, it should stop after processing the actual number
of filter items only.

Thanks,
Namhyung


> >
> > Namhyung Kim (7):
> >   perf bpf filter: Introduce basic BPF filter expression
> >   perf bpf filter: Implement event sample filtering
> >   perf record: Add BPF event filter support
> >   perf record: Record dropped sample count
> >   perf bpf filter: Add 'pid' sample data support
> >   perf bpf filter: Add more weight sample data support
> >   perf bpf filter: Add data_src sample data support
> >
> >  tools/perf/Documentation/perf-record.txt     |  10 +-
> >  tools/perf/Makefile.perf                     |   2 +-
> >  tools/perf/builtin-record.c                  |  46 ++++--
> >  tools/perf/util/Build                        |  16 ++
> >  tools/perf/util/bpf-filter.c                 | 117 ++++++++++++++
> >  tools/perf/util/bpf-filter.h                 |  48 ++++++
> >  tools/perf/util/bpf-filter.l                 | 146 ++++++++++++++++++
> >  tools/perf/util/bpf-filter.y                 |  55 +++++++
> >  tools/perf/util/bpf_counter.c                |   3 +-
> >  tools/perf/util/bpf_skel/sample-filter.h     |  25 +++
> >  tools/perf/util/bpf_skel/sample_filter.bpf.c | 152 +++++++++++++++++++
> >  tools/perf/util/evsel.c                      |   2 +
> >  tools/perf/util/evsel.h                      |   7 +-
> >  tools/perf/util/parse-events.c               |   4 +
> >  tools/perf/util/session.c                    |   3 +-
> >  15 files changed, 615 insertions(+), 21 deletions(-)
> >  create mode 100644 tools/perf/util/bpf-filter.c
> >  create mode 100644 tools/perf/util/bpf-filter.h
> >  create mode 100644 tools/perf/util/bpf-filter.l
> >  create mode 100644 tools/perf/util/bpf-filter.y
> >  create mode 100644 tools/perf/util/bpf_skel/sample-filter.h
> >  create mode 100644 tools/perf/util/bpf_skel/sample_filter.bpf.c
> >
> >
> > base-commit: 37f322cd58d81a9d46456531281c908de9ef6e42
> > --
> > 2.39.1.581.gbfd45094c4-goog
> >
Arnaldo Carvalho de Melo Feb. 14, 2023, 7:16 p.m. UTC | #3
Em Tue, Feb 14, 2023 at 08:57:58AM -0800, Ian Rogers escreveu:
> On Mon, Feb 13, 2023 at 9:05 PM Namhyung Kim <namhyung@kernel.org> wrote:
> > There have been requests for more sophisticated perf event sample
> > filtering based on the sample data.  Recently the kernel added BPF
> > programs can access perf sample data and this is the userspace part
> > to enable such a filtering.

> > This still has some rough edges and needs more improvements.  But
> > I'd like to share the current work and get some feedback for the
> > directions and idea for further improvements.

> > The kernel changes are in the tip.git tree (perf/core branch) for now.
> > perf record has --filter option to set filters on the last specified
> > event in the command line.  It worked only for tracepoints and Intel
> > PT events so far.  This patchset extends it to have 'bpf:' prefix in
> > order to enable the general sample filters using BPF for any events.

> > A new filter expression parser was added (using flex/bison) to process
> > the filter string.  Right now, it only accepts very simple expressions
> > separated by comma.  I'd like to keep the filter expression as simple
> > as possible.

> > It requires samples satisfy all the filter expressions otherwise it'd
> > drop the sample.  IOW filter expressions are connected with logical AND
> > operations implicitly.

> > Essentially the BPF filter expression is:

> >   "bpf:" <term> <operator> <value> ("," <term> <operator> <value>)*

bpf is the technology used for that, but this really is about filtering
by fields in the sample type, right? So perhaps we could remove that
"bpf:" part and simply do:

  sudo ./perf record -e cycles --filter 'period > 10000' -- ./perf test -w noploop

And notice that this requires this new mechanism and just use it? It
gets more compact and should be unambiguous for non-tracepoint events?

And for tracepoint events if we can use both mechanisms, then use the
tracepoint one since it requires less setup?

Perhaps use "sample_type.field" to disambiguate if we would like to get a
field from the sample_type and another in the tracepoint if both have
the same name?

And how difficult it would be to just accept the same syntax (or a
superset) of what is available for tracepoint filters? I.e. allow || as
well as &&.

Great stuff!

- Arnaldo

> > The <term> can be one of:
> >   ip, id, tid, pid, cpu, time, addr, period, txn, weight, phys_addr,
> >   code_pgsz, data_pgsz, weight1, weight2, weight3, ins_lat, retire_lat,
> >   p_stage_cyc, mem_op, mem_lvl, mem_snoop, mem_remote, mem_lock,
> >   mem_dtlb, mem_blk, mem_hops
> >
> > The <operator> can be one of:
> >   ==, !=, >, >=, <, <=, &
> >
> > The <value> can be one of:
> >   <number> (for any term)
> >   na, load, store, pfetch, exec (for mem_op)
> >   l1, l2, l3, l4, cxl, io, any_cache, lfb, ram, pmem (for mem_lvl)
> >   na, none, hit, miss, hitm, fwd, peer (for mem_snoop)
> >   remote (for mem_remote)
> >   na, locked (for mem_locked)
> >   na, l1_hit, l1_miss, l2_hit, l2_miss, any_hit, any_miss, walk, fault (for mem_dtlb)
> >   na, by_data, by_addr (for mem_blk)
> >   hops0, hops1, hops2, hops3 (for mem_hops)
> >
> > I plan to improve it with range expressions like for ip or addr and it
> > should support symbols like the existing addr-filters.  Also cgroup
> > should understand and convert cgroup names to IDs.
> >
> > Let's take a look at some examples.  The following is to profile a user
> > program on the command line.  When the frequency mode is used, it starts
> > with a very small period (i.e. 1) and adjust it on every interrupt (NMI)
> > to catch up the given frequency.
> >
> >   $ ./perf record -- ./perf test -w noploop
> >   [ perf record: Woken up 1 times to write data ]
> >   [ perf record: Captured and wrote 0.263 MB perf.data (4006 samples) ]
> >
> >   $ ./perf script -F pid,period,event,ip,sym | head
> >   36695          1 cycles:  ffffffffbab12ddd perf_event_exec
> >   36695          1 cycles:  ffffffffbab12ddd perf_event_exec
> >   36695          5 cycles:  ffffffffbab12ddd perf_event_exec
> >   36695         46 cycles:  ffffffffbab12de5 perf_event_exec
> >   36695       1163 cycles:  ffffffffba80a0eb x86_pmu_disable_all
> >   36695       1304 cycles:  ffffffffbaa19507 __hrtimer_get_next_event
> >   36695       8143 cycles:  ffffffffbaa186f9 __run_timers
> >   36695      69040 cycles:  ffffffffbaa0c393 rcu_segcblist_ready_cbs
> >   36695     355117 cycles:            4b0da4 noploop
> >   36695     321861 cycles:            4b0da4 noploop
> >
> > If you want to skip the first few samples that have small periods, you
> > can do like this (note it requires root due to BPF).
> >
> >   $ sudo ./perf record -e cycles --filter 'bpf: period > 10000' -- ./perf test -w noploop
> >   [ perf record: Woken up 1 times to write data ]
> >   [ perf record: Captured and wrote 0.262 MB perf.data (3990 samples) ]
> >
> >   $ sudo ./perf script -F pid,period,event,ip,sym | head
> >   39524      58253 cycles:  ffffffffba97dac0 update_rq_clock
> >   39524     232657 cycles:            4b0da2 noploop
> >   39524     210981 cycles:            4b0da2 noploop
> >   39524     282882 cycles:            4b0da4 noploop
> >   39524     392180 cycles:            4b0da4 noploop
> >   39524     456058 cycles:            4b0da4 noploop
> >   39524     415196 cycles:            4b0da2 noploop
> >   39524     462721 cycles:            4b0da4 noploop
> >   39524     526272 cycles:            4b0da2 noploop
> >   39524     565569 cycles:            4b0da4 noploop
> >
> > Maybe more useful example is when it deals with precise memory events.
> > On AMD processors with IBS, you can filter only memory load with L1
> > dTLB is missed like below.
> >
> >   $ sudo ./perf record -ad -e ibs_op//p \
> >   > --filter 'bpf: mem_op == load, mem_dtlb > l1_hit' sleep 1
> >   [ perf record: Woken up 1 times to write data ]
> >   [ perf record: Captured and wrote 1.338 MB perf.data (15 samples) ]
> >
> >   $ sudo ./perf script -F data_src | head
> >           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> >           49080142 |OP LOAD|LVL L1 hit|SNP N/A|TLB L2 hit|LCK N/A|BLK  N/A
> >           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> >           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> >           51088842 |OP LOAD|LVL L3 or Remote Cache (1 hop) hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> >           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> >           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> >           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> >           49080442 |OP LOAD|LVL L2 hit|SNP N/A|TLB L2 hit|LCK N/A|BLK  N/A
> >           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> >
> > You can also check the number of dropped samples in LOST_SAMPLES events
> > using perf report --stat command.
> >
> >   $ sudo ./perf report --stat
> >
> >   Aggregated stats:
> >              TOTAL events:      16066
> >               MMAP events:         22  ( 0.1%)
> >               COMM events:       4166  (25.9%)
> >               EXIT events:          1  ( 0.0%)
> >           THROTTLE events:        816  ( 5.1%)
> >         UNTHROTTLE events:        613  ( 3.8%)
> >               FORK events:       4165  (25.9%)
> >             SAMPLE events:         15  ( 0.1%)
> >              MMAP2 events:       6133  (38.2%)
> >       LOST_SAMPLES events:          1  ( 0.0%)
> >            KSYMBOL events:         69  ( 0.4%)
> >          BPF_EVENT events:         57  ( 0.4%)
> >     FINISHED_ROUND events:          3  ( 0.0%)
> >           ID_INDEX events:          1  ( 0.0%)
> >         THREAD_MAP events:          1  ( 0.0%)
> >            CPU_MAP events:          1  ( 0.0%)
> >          TIME_CONV events:          1  ( 0.0%)
> >      FINISHED_INIT events:          1  ( 0.0%)
> >   ibs_op//p stats:
> >             SAMPLE events:         15
> >       LOST_SAMPLES events:       3991
> >
> > Note that the total aggregated stats show 1 LOST_SAMPLES event but
> > per event stats show 3991 events because it's the actual number of
> > dropped samples while the aggregated stats has the number of record.
> > Maybe we need to change the per-event stats to 'LOST_SAMPLES count'
> > to avoid the confusion.
> >
> > The code is available at 'perf/bpf-filter-v1' branch in my tree.
> >
> >   git://git.kernel.org/pub/scm/linux/kernel/git/namhyung/linux-perf.git
> >
> > Again, you need tip/perf/core kernel for this to work.
> > Any feedback is welcome.
> 
> This is great! I wonder about related clean up:
> 
>  - can we remove BPF events as this is a better feature?
>    - I believe BPF events are flaky, seldom used (with the exception
> of the augmented syscalls for perf trace, which really should move to
> a BPF skeleton as most people don't know how to use it) and they add a
> bunch of complexity. A particular complexity I care about is that the
> path separator forward slash ('/') is also the modifier separator for
> events.
> 
>  - what will happen with multiple events/metrics? Perhaps there should
> be a way of listing filters so that each filter applies to the
> appropriate event in the event list, like cgroups and -G. For metrics
> we shuffle the list of events and so maybe the filters need some way
> to specify which event they apply to.
> 
>  - It feels like there should be some BPF way of overcoming the fixed
> length number of filters so it is still bounded but not a hardcoded
> number.
> 
> Thanks,
> Ian
> 
> 
> > Thanks,
> > Namhyung
> >
> > Namhyung Kim (7):
> >   perf bpf filter: Introduce basic BPF filter expression
> >   perf bpf filter: Implement event sample filtering
> >   perf record: Add BPF event filter support
> >   perf record: Record dropped sample count
> >   perf bpf filter: Add 'pid' sample data support
> >   perf bpf filter: Add more weight sample data support
> >   perf bpf filter: Add data_src sample data support
> >
> >  tools/perf/Documentation/perf-record.txt     |  10 +-
> >  tools/perf/Makefile.perf                     |   2 +-
> >  tools/perf/builtin-record.c                  |  46 ++++--
> >  tools/perf/util/Build                        |  16 ++
> >  tools/perf/util/bpf-filter.c                 | 117 ++++++++++++++
> >  tools/perf/util/bpf-filter.h                 |  48 ++++++
> >  tools/perf/util/bpf-filter.l                 | 146 ++++++++++++++++++
> >  tools/perf/util/bpf-filter.y                 |  55 +++++++
> >  tools/perf/util/bpf_counter.c                |   3 +-
> >  tools/perf/util/bpf_skel/sample-filter.h     |  25 +++
> >  tools/perf/util/bpf_skel/sample_filter.bpf.c | 152 +++++++++++++++++++
> >  tools/perf/util/evsel.c                      |   2 +
> >  tools/perf/util/evsel.h                      |   7 +-
> >  tools/perf/util/parse-events.c               |   4 +
> >  tools/perf/util/session.c                    |   3 +-
> >  15 files changed, 615 insertions(+), 21 deletions(-)
> >  create mode 100644 tools/perf/util/bpf-filter.c
> >  create mode 100644 tools/perf/util/bpf-filter.h
> >  create mode 100644 tools/perf/util/bpf-filter.l
> >  create mode 100644 tools/perf/util/bpf-filter.y
> >  create mode 100644 tools/perf/util/bpf_skel/sample-filter.h
> >  create mode 100644 tools/perf/util/bpf_skel/sample_filter.bpf.c
> >
> >
> > base-commit: 37f322cd58d81a9d46456531281c908de9ef6e42
> > --
> > 2.39.1.581.gbfd45094c4-goog
> >
Namhyung Kim Feb. 14, 2023, 10:16 p.m. UTC | #4
Hi Arnaldo,

On Tue, Feb 14, 2023 at 11:16 AM Arnaldo Carvalho de Melo
<acme@kernel.org> wrote:
>
> Em Tue, Feb 14, 2023 at 08:57:58AM -0800, Ian Rogers escreveu:
> > On Mon, Feb 13, 2023 at 9:05 PM Namhyung Kim <namhyung@kernel.org> wrote:
> > > There have been requests for more sophisticated perf event sample
> > > filtering based on the sample data.  Recently the kernel added BPF
> > > programs can access perf sample data and this is the userspace part
> > > to enable such a filtering.
>
> > > This still has some rough edges and needs more improvements.  But
> > > I'd like to share the current work and get some feedback for the
> > > directions and idea for further improvements.
>
> > > The kernel changes are in the tip.git tree (perf/core branch) for now.
> > > perf record has --filter option to set filters on the last specified
> > > event in the command line.  It worked only for tracepoints and Intel
> > > PT events so far.  This patchset extends it to have 'bpf:' prefix in
> > > order to enable the general sample filters using BPF for any events.
>
> > > A new filter expression parser was added (using flex/bison) to process
> > > the filter string.  Right now, it only accepts very simple expressions
> > > separated by comma.  I'd like to keep the filter expression as simple
> > > as possible.
>
> > > It requires samples satisfy all the filter expressions otherwise it'd
> > > drop the sample.  IOW filter expressions are connected with logical AND
> > > operations implicitly.
>
> > > Essentially the BPF filter expression is:
>
> > >   "bpf:" <term> <operator> <value> ("," <term> <operator> <value>)*
>
> bpf is the technology used for that, but this really is about filtering
> by fields in the sample type, right? So perhaps we could remove that
> "bpf:" part and simply do:
>
>   sudo ./perf record -e cycles --filter 'period > 10000' -- ./perf test -w noploop
>
> And notice that this requires this new mechanism and just use it? It
> gets more compact and should be unambiguous for non-tracepoint events?
>
> And for tracepoint events if we can use both mechanisms, then use the
> tracepoint one since it requires less setup?

Sure, it'd work if we could select the filter mechanism based on the
event type.  One thing to note is BPF filter requires root permission
even if the event itself does not.  Users might be surprised if it
suddenly requires root for their userspace profiling.

>
> Perhaps use "sample_type.field" to disambiguate if we would like to get a
> field from the sample_type and another in the tracepoint if both have
> the same name?

I think the tracepoint filters are different as they work on the event-
specific data field.  From the sample data's perspective, it's just
RAW data and current BPF filters do nothing with it.

So I'd rather simply delegate it to the tracepoint.

>
> And how difficult it would be to just accept the same syntax (or a
> superset) of what is available for tracepoint filters? I.e. allow || as
> well as &&.

Making the parser accept those syntax would not be that difficult.
But I'm afraid of the BPF program doing the job and how we can
build the map to achieve that.

>
> Great stuff!

Thanks!
Namhyung


>
> > > The <term> can be one of:
> > >   ip, id, tid, pid, cpu, time, addr, period, txn, weight, phys_addr,
> > >   code_pgsz, data_pgsz, weight1, weight2, weight3, ins_lat, retire_lat,
> > >   p_stage_cyc, mem_op, mem_lvl, mem_snoop, mem_remote, mem_lock,
> > >   mem_dtlb, mem_blk, mem_hops
> > >
> > > The <operator> can be one of:
> > >   ==, !=, >, >=, <, <=, &
> > >
> > > The <value> can be one of:
> > >   <number> (for any term)
> > >   na, load, store, pfetch, exec (for mem_op)
> > >   l1, l2, l3, l4, cxl, io, any_cache, lfb, ram, pmem (for mem_lvl)
> > >   na, none, hit, miss, hitm, fwd, peer (for mem_snoop)
> > >   remote (for mem_remote)
> > >   na, locked (for mem_locked)
> > >   na, l1_hit, l1_miss, l2_hit, l2_miss, any_hit, any_miss, walk, fault (for mem_dtlb)
> > >   na, by_data, by_addr (for mem_blk)
> > >   hops0, hops1, hops2, hops3 (for mem_hops)
> > >
> > > I plan to improve it with range expressions like for ip or addr and it
> > > should support symbols like the existing addr-filters.  Also cgroup
> > > should understand and convert cgroup names to IDs.
> > >
> > > Let's take a look at some examples.  The following is to profile a user
> > > program on the command line.  When the frequency mode is used, it starts
> > > with a very small period (i.e. 1) and adjust it on every interrupt (NMI)
> > > to catch up the given frequency.
> > >
> > >   $ ./perf record -- ./perf test -w noploop
> > >   [ perf record: Woken up 1 times to write data ]
> > >   [ perf record: Captured and wrote 0.263 MB perf.data (4006 samples) ]
> > >
> > >   $ ./perf script -F pid,period,event,ip,sym | head
> > >   36695          1 cycles:  ffffffffbab12ddd perf_event_exec
> > >   36695          1 cycles:  ffffffffbab12ddd perf_event_exec
> > >   36695          5 cycles:  ffffffffbab12ddd perf_event_exec
> > >   36695         46 cycles:  ffffffffbab12de5 perf_event_exec
> > >   36695       1163 cycles:  ffffffffba80a0eb x86_pmu_disable_all
> > >   36695       1304 cycles:  ffffffffbaa19507 __hrtimer_get_next_event
> > >   36695       8143 cycles:  ffffffffbaa186f9 __run_timers
> > >   36695      69040 cycles:  ffffffffbaa0c393 rcu_segcblist_ready_cbs
> > >   36695     355117 cycles:            4b0da4 noploop
> > >   36695     321861 cycles:            4b0da4 noploop
> > >
> > > If you want to skip the first few samples that have small periods, you
> > > can do like this (note it requires root due to BPF).
> > >
> > >   $ sudo ./perf record -e cycles --filter 'bpf: period > 10000' -- ./perf test -w noploop
> > >   [ perf record: Woken up 1 times to write data ]
> > >   [ perf record: Captured and wrote 0.262 MB perf.data (3990 samples) ]
> > >
> > >   $ sudo ./perf script -F pid,period,event,ip,sym | head
> > >   39524      58253 cycles:  ffffffffba97dac0 update_rq_clock
> > >   39524     232657 cycles:            4b0da2 noploop
> > >   39524     210981 cycles:            4b0da2 noploop
> > >   39524     282882 cycles:            4b0da4 noploop
> > >   39524     392180 cycles:            4b0da4 noploop
> > >   39524     456058 cycles:            4b0da4 noploop
> > >   39524     415196 cycles:            4b0da2 noploop
> > >   39524     462721 cycles:            4b0da4 noploop
> > >   39524     526272 cycles:            4b0da2 noploop
> > >   39524     565569 cycles:            4b0da4 noploop
> > >
> > > Maybe more useful example is when it deals with precise memory events.
> > > On AMD processors with IBS, you can filter only memory load with L1
> > > dTLB is missed like below.
> > >
> > >   $ sudo ./perf record -ad -e ibs_op//p \
> > >   > --filter 'bpf: mem_op == load, mem_dtlb > l1_hit' sleep 1
> > >   [ perf record: Woken up 1 times to write data ]
> > >   [ perf record: Captured and wrote 1.338 MB perf.data (15 samples) ]
> > >
> > >   $ sudo ./perf script -F data_src | head
> > >           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> > >           49080142 |OP LOAD|LVL L1 hit|SNP N/A|TLB L2 hit|LCK N/A|BLK  N/A
> > >           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> > >           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> > >           51088842 |OP LOAD|LVL L3 or Remote Cache (1 hop) hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> > >           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> > >           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> > >           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> > >           49080442 |OP LOAD|LVL L2 hit|SNP N/A|TLB L2 hit|LCK N/A|BLK  N/A
> > >           51080242 |OP LOAD|LVL LFB/MAB hit|SNP N/A|TLB L2 miss|LCK N/A|BLK  N/A
> > >
> > > You can also check the number of dropped samples in LOST_SAMPLES events
> > > using perf report --stat command.
> > >
> > >   $ sudo ./perf report --stat
> > >
> > >   Aggregated stats:
> > >              TOTAL events:      16066
> > >               MMAP events:         22  ( 0.1%)
> > >               COMM events:       4166  (25.9%)
> > >               EXIT events:          1  ( 0.0%)
> > >           THROTTLE events:        816  ( 5.1%)
> > >         UNTHROTTLE events:        613  ( 3.8%)
> > >               FORK events:       4165  (25.9%)
> > >             SAMPLE events:         15  ( 0.1%)
> > >              MMAP2 events:       6133  (38.2%)
> > >       LOST_SAMPLES events:          1  ( 0.0%)
> > >            KSYMBOL events:         69  ( 0.4%)
> > >          BPF_EVENT events:         57  ( 0.4%)
> > >     FINISHED_ROUND events:          3  ( 0.0%)
> > >           ID_INDEX events:          1  ( 0.0%)
> > >         THREAD_MAP events:          1  ( 0.0%)
> > >            CPU_MAP events:          1  ( 0.0%)
> > >          TIME_CONV events:          1  ( 0.0%)
> > >      FINISHED_INIT events:          1  ( 0.0%)
> > >   ibs_op//p stats:
> > >             SAMPLE events:         15
> > >       LOST_SAMPLES events:       3991
> > >
> > > Note that the total aggregated stats show 1 LOST_SAMPLES event but
> > > per event stats show 3991 events because it's the actual number of
> > > dropped samples while the aggregated stats has the number of record.
> > > Maybe we need to change the per-event stats to 'LOST_SAMPLES count'
> > > to avoid the confusion.
> > >
> > > The code is available at 'perf/bpf-filter-v1' branch in my tree.
> > >
> > >   git://git.kernel.org/pub/scm/linux/kernel/git/namhyung/linux-perf.git
> > >
> > > Again, you need tip/perf/core kernel for this to work.
> > > Any feedback is welcome.
> >
> > This is great! I wonder about related clean up:
> >
> >  - can we remove BPF events as this is a better feature?
> >    - I believe BPF events are flaky, seldom used (with the exception
> > of the augmented syscalls for perf trace, which really should move to
> > a BPF skeleton as most people don't know how to use it) and they add a
> > bunch of complexity. A particular complexity I care about is that the
> > path separator forward slash ('/') is also the modifier separator for
> > events.
> >
> >  - what will happen with multiple events/metrics? Perhaps there should
> > be a way of listing filters so that each filter applies to the
> > appropriate event in the event list, like cgroups and -G. For metrics
> > we shuffle the list of events and so maybe the filters need some way
> > to specify which event they apply to.
> >
> >  - It feels like there should be some BPF way of overcoming the fixed
> > length number of filters so it is still bounded but not a hardcoded
> > number.
> >
> > Thanks,
> > Ian
> >
> >
> > > Thanks,
> > > Namhyung
> > >
> > > Namhyung Kim (7):
> > >   perf bpf filter: Introduce basic BPF filter expression
> > >   perf bpf filter: Implement event sample filtering
> > >   perf record: Add BPF event filter support
> > >   perf record: Record dropped sample count
> > >   perf bpf filter: Add 'pid' sample data support
> > >   perf bpf filter: Add more weight sample data support
> > >   perf bpf filter: Add data_src sample data support
> > >
> > >  tools/perf/Documentation/perf-record.txt     |  10 +-
> > >  tools/perf/Makefile.perf                     |   2 +-
> > >  tools/perf/builtin-record.c                  |  46 ++++--
> > >  tools/perf/util/Build                        |  16 ++
> > >  tools/perf/util/bpf-filter.c                 | 117 ++++++++++++++
> > >  tools/perf/util/bpf-filter.h                 |  48 ++++++
> > >  tools/perf/util/bpf-filter.l                 | 146 ++++++++++++++++++
> > >  tools/perf/util/bpf-filter.y                 |  55 +++++++
> > >  tools/perf/util/bpf_counter.c                |   3 +-
> > >  tools/perf/util/bpf_skel/sample-filter.h     |  25 +++
> > >  tools/perf/util/bpf_skel/sample_filter.bpf.c | 152 +++++++++++++++++++
> > >  tools/perf/util/evsel.c                      |   2 +
> > >  tools/perf/util/evsel.h                      |   7 +-
> > >  tools/perf/util/parse-events.c               |   4 +
> > >  tools/perf/util/session.c                    |   3 +-
> > >  15 files changed, 615 insertions(+), 21 deletions(-)
> > >  create mode 100644 tools/perf/util/bpf-filter.c
> > >  create mode 100644 tools/perf/util/bpf-filter.h
> > >  create mode 100644 tools/perf/util/bpf-filter.l
> > >  create mode 100644 tools/perf/util/bpf-filter.y
> > >  create mode 100644 tools/perf/util/bpf_skel/sample-filter.h
> > >  create mode 100644 tools/perf/util/bpf_skel/sample_filter.bpf.c
> > >
> > >
> > > base-commit: 37f322cd58d81a9d46456531281c908de9ef6e42
> > > --
> > > 2.39.1.581.gbfd45094c4-goog
> > >
>
> --
>
> - Arnaldo
Jiri Olsa Feb. 21, 2023, 11:54 a.m. UTC | #5
On Tue, Feb 14, 2023 at 10:01:41AM -0800, Namhyung Kim wrote:
> Hi Ian,
> 
> On Tue, Feb 14, 2023 at 8:58 AM Ian Rogers <irogers@google.com> wrote:
> >
> > On Mon, Feb 13, 2023 at 9:05 PM Namhyung Kim <namhyung@kernel.org> wrote:
> > >
> > > Hello,
> > >
> > > There have been requests for more sophisticated perf event sample
> > > filtering based on the sample data.  Recently the kernel added BPF
> > > programs can access perf sample data and this is the userspace part
> > > to enable such a filtering.
> > >
> > > This still has some rough edges and needs more improvements.  But
> > > I'd like to share the current work and get some feedback for the
> > > directions and idea for further improvements.
> > >
> > > The kernel changes are in the tip.git tree (perf/core branch) for now.
> > > perf record has --filter option to set filters on the last specified
> > > event in the command line.  It worked only for tracepoints and Intel
> > > PT events so far.  This patchset extends it to have 'bpf:' prefix in
> > > order to enable the general sample filters using BPF for any events.
> > >
> > > A new filter expression parser was added (using flex/bison) to process
> > > the filter string.  Right now, it only accepts very simple expressions
> > > separated by comma.  I'd like to keep the filter expression as simple
> > > as possible.
> > >
> > > It requires samples satisfy all the filter expressions otherwise it'd
> > > drop the sample.  IOW filter expressions are connected with logical AND
> > > operations implicitly.
> > >
> > > Essentially the BPF filter expression is:
> > >
> > >   "bpf:" <term> <operator> <value> ("," <term> <operator> <value>)*
> > >
> > > The <term> can be one of:
> > >   ip, id, tid, pid, cpu, time, addr, period, txn, weight, phys_addr,
> > >   code_pgsz, data_pgsz, weight1, weight2, weight3, ins_lat, retire_lat,
> > >   p_stage_cyc, mem_op, mem_lvl, mem_snoop, mem_remote, mem_lock,
> > >   mem_dtlb, mem_blk, mem_hops
> > >
> > > The <operator> can be one of:
> > >   ==, !=, >, >=, <, <=, &
> > >
> > > The <value> can be one of:
> > >   <number> (for any term)
> > >   na, load, store, pfetch, exec (for mem_op)
> > >   l1, l2, l3, l4, cxl, io, any_cache, lfb, ram, pmem (for mem_lvl)
> > >   na, none, hit, miss, hitm, fwd, peer (for mem_snoop)
> > >   remote (for mem_remote)
> > >   na, locked (for mem_locked)
> > >   na, l1_hit, l1_miss, l2_hit, l2_miss, any_hit, any_miss, walk, fault (for mem_dtlb)
> > >   na, by_data, by_addr (for mem_blk)
> > >   hops0, hops1, hops2, hops3 (for mem_hops)
> > >
> > > I plan to improve it with range expressions like for ip or addr and it
> > > should support symbols like the existing addr-filters.  Also cgroup
> > > should understand and convert cgroup names to IDs.

this seems similar to what ftrace is doing in filter_match_preds,
I checked the code briefly and I wonder if we shoud be able to write
that function logic in bpf, assuming that the filter is prepared in
user space

it might solve the 'part' data problem in generic way.. but I might be
missing some blocker of course.. just an idea ;-)

could replace the tracepoint filters.. if we actually care

SNIP

> > > Note that the total aggregated stats show 1 LOST_SAMPLES event but
> > > per event stats show 3991 events because it's the actual number of
> > > dropped samples while the aggregated stats has the number of record.
> > > Maybe we need to change the per-event stats to 'LOST_SAMPLES count'
> > > to avoid the confusion.
> > >
> > > The code is available at 'perf/bpf-filter-v1' branch in my tree.
> > >
> > >   git://git.kernel.org/pub/scm/linux/kernel/git/namhyung/linux-perf.git
> > >
> > > Again, you need tip/perf/core kernel for this to work.
> > > Any feedback is welcome.
> >
> > This is great! I wonder about related clean up:

+1

> >
> >  - can we remove BPF events as this is a better feature?
> >    - I believe BPF events are flaky, seldom used (with the exception
> > of the augmented syscalls for perf trace, which really should move to
> > a BPF skeleton as most people don't know how to use it) and they add a
> > bunch of complexity. A particular complexity I care about is that the
> > path separator forward slash ('/') is also the modifier separator for
> > events.
> 
> Well.. I actually never tried the BPF events myself :)
> I think we can deprecate it and get rid of it once the perf trace
> conversion is done.

+1 ;-) would be awesome

jirka
Namhyung Kim Feb. 22, 2023, 7:42 p.m. UTC | #6
Hi Jiri,

On Tue, Feb 21, 2023 at 3:54 AM Jiri Olsa <olsajiri@gmail.com> wrote:
>
> On Tue, Feb 14, 2023 at 10:01:41AM -0800, Namhyung Kim wrote:
> > Hi Ian,
> >
> > On Tue, Feb 14, 2023 at 8:58 AM Ian Rogers <irogers@google.com> wrote:
> > >
> > > On Mon, Feb 13, 2023 at 9:05 PM Namhyung Kim <namhyung@kernel.org> wrote:
> > > >
> > > > Hello,
> > > >
> > > > There have been requests for more sophisticated perf event sample
> > > > filtering based on the sample data.  Recently the kernel added BPF
> > > > programs can access perf sample data and this is the userspace part
> > > > to enable such a filtering.
> > > >
> > > > This still has some rough edges and needs more improvements.  But
> > > > I'd like to share the current work and get some feedback for the
> > > > directions and idea for further improvements.
> > > >
> > > > The kernel changes are in the tip.git tree (perf/core branch) for now.
> > > > perf record has --filter option to set filters on the last specified
> > > > event in the command line.  It worked only for tracepoints and Intel
> > > > PT events so far.  This patchset extends it to have 'bpf:' prefix in
> > > > order to enable the general sample filters using BPF for any events.
> > > >
> > > > A new filter expression parser was added (using flex/bison) to process
> > > > the filter string.  Right now, it only accepts very simple expressions
> > > > separated by comma.  I'd like to keep the filter expression as simple
> > > > as possible.
> > > >
> > > > It requires samples satisfy all the filter expressions otherwise it'd
> > > > drop the sample.  IOW filter expressions are connected with logical AND
> > > > operations implicitly.
> > > >
> > > > Essentially the BPF filter expression is:
> > > >
> > > >   "bpf:" <term> <operator> <value> ("," <term> <operator> <value>)*
> > > >
> > > > The <term> can be one of:
> > > >   ip, id, tid, pid, cpu, time, addr, period, txn, weight, phys_addr,
> > > >   code_pgsz, data_pgsz, weight1, weight2, weight3, ins_lat, retire_lat,
> > > >   p_stage_cyc, mem_op, mem_lvl, mem_snoop, mem_remote, mem_lock,
> > > >   mem_dtlb, mem_blk, mem_hops
> > > >
> > > > The <operator> can be one of:
> > > >   ==, !=, >, >=, <, <=, &
> > > >
> > > > The <value> can be one of:
> > > >   <number> (for any term)
> > > >   na, load, store, pfetch, exec (for mem_op)
> > > >   l1, l2, l3, l4, cxl, io, any_cache, lfb, ram, pmem (for mem_lvl)
> > > >   na, none, hit, miss, hitm, fwd, peer (for mem_snoop)
> > > >   remote (for mem_remote)
> > > >   na, locked (for mem_locked)
> > > >   na, l1_hit, l1_miss, l2_hit, l2_miss, any_hit, any_miss, walk, fault (for mem_dtlb)
> > > >   na, by_data, by_addr (for mem_blk)
> > > >   hops0, hops1, hops2, hops3 (for mem_hops)
> > > >
> > > > I plan to improve it with range expressions like for ip or addr and it
> > > > should support symbols like the existing addr-filters.  Also cgroup
> > > > should understand and convert cgroup names to IDs.
>
> this seems similar to what ftrace is doing in filter_match_preds,
> I checked the code briefly and I wonder if we shoud be able to write
> that function logic in bpf, assuming that the filter is prepared in
> user space
>
> it might solve the 'part' data problem in generic way.. but I might be
> missing some blocker of course.. just an idea ;-)
>
> could replace the tracepoint filters.. if we actually care

I'm not sure about replacing tracepoint filters.  IIRC BPF is optional,
then tracepoints should work without it.  From the BPF's perspective,
it has its own way of handling tracepoints so no need to deal with
perf or event tracing (ftrace) for that.

From the perf's perspective, I think it can use either the existing ftrace
filters or build a new BPF filter for each event.  But it cannot use BTF
for perf tracepoint events at least for now.  Certainly it can use RAW
sample data and parse the event format to access the fields but I'm
not sure it's worth doing that. :)

Thanks,
Namhyung