On Wed, Dec 22, 2021 at 2:17 PM Andrii Nakryiko
<andrii.nakryiko@xxxxxxxxx> wrote:
>
> On Wed, Dec 22, 2021 at 5:44 AM Jiri Olsa <jolsa@xxxxxxxxxx> wrote:
> >
> > On Tue, Dec 21, 2021 at 01:58:14PM -0800, Andrii Nakryiko wrote:
> > > On Tue, Dec 21, 2021 at 12:23 AM Jiri Olsa <jolsa@xxxxxxxxxx> wrote:
> > > >
> > > > On Thu, Dec 16, 2021 at 02:21:08PM -0800, Christy Lee wrote:
> > > > > bpf__object_next is deprecated, track bpf_objects directly in
> > > > > perf instead.
> > > > >
> > > > > Signed-off-by: Christy Lee <christylee@xxxxxx>
> > > > > Acked-by: Andrii Nakryiko <andrii@xxxxxxxxxx>
> > > > > ---
> > > > > tools/perf/util/bpf-loader.c | 72 +++++++++++++++++++++++++++---------
> > > > > tools/perf/util/bpf-loader.h | 1 +
> > > > > 2 files changed, 55 insertions(+), 18 deletions(-)
> > > > >
> > > > > diff --git a/tools/perf/util/bpf-loader.c b/tools/perf/util/bpf-loader.c
> > > > > index 528aeb0ab79d..9e3988fd719a 100644
> > > > > --- a/tools/perf/util/bpf-loader.c
> > > > > +++ b/tools/perf/util/bpf-loader.c
> > > > > @@ -29,9 +29,6 @@
> > > > >
> > > > > #include <internal/xyarray.h>
> > > > >
> > > > > -/* temporarily disable libbpf deprecation warnings */
> > > > > -#pragma GCC diagnostic ignored "-Wdeprecated-declarations"
> > > > > -
> > > > > static int libbpf_perf_print(enum libbpf_print_level level __attribute__((unused)),
> > > > > const char *fmt, va_list args)
> > > > > {
> > > > > @@ -49,6 +46,36 @@ struct bpf_prog_priv {
> > > > > int *type_mapping;
> > > > > };
> > > > >
> > > > > +struct bpf_perf_object {
> > > > > + struct bpf_object *obj;
> > > > > + struct list_head list;
> > > > > +};
> > > > > +
> > > > > +static LIST_HEAD(bpf_objects_list);
> > > >
> > > > hum, so this duplicates libbpf's bpf_objects_list,
> > > > how do objects get on this list?
> > >
> > > yep, this list needs to be updated on perf side each time
> > > bpf_object__open() (and any variant of open) is called.
> > >
> > > >
> > > > could you please put more comments in changelog
> > > > and share how you tested this?
> > >
> > > I actually have no idea how to test this as well, can you please share
> > > some ideas?
> > >
> >
> > I don't use it, I just know it's there.. that's why I asked ;-)
> >
> > it's possible to specify bpf program on the perf command line
> > to be attached to event, like:
> >
> > # cat tools/perf/examples/bpf/hello.c
> > #include <stdio.h>
> >
> > int syscall_enter(openat)(void *args)
> > {
> > puts("Hello, world\n");
> > return 0;
> > }
> >
> > license(GPL);
> > #
> > # perf trace -e openat,tools/perf/examples/bpf/hello.c cat /etc/passwd > /dev/null
> > 0.016 ( ): __bpf_stdout__:Hello, world
> > 0.018 ( 0.010 ms): cat/9079 openat(dfd: CWD, filename: /etc/ld.so.cache, flags: CLOEXEC) = 3
> > 0.057 ( ): __bpf_stdout__:Hello, world
> > 0.059 ( 0.011 ms): cat/9079 openat(dfd: CWD, filename: /lib64/libc.so.6, flags: CLOEXEC) = 3
> > 0.417 ( ): __bpf_stdout__:Hello, world
> > 0.419 ( 0.009 ms): cat/9079 openat(dfd: CWD, filename: /etc/passwd) = 3
> > #
> >
> > I took that example from commit message
[...]
I found the original commit aa3abf30bb28addcf593578d37447d42e3f65fc3
that included a test case, but I'm having trouble reproducing it due to syntax
error. I am running this on bpf-next master without my patches.
I ran 'perf test -v LLVM' and used it's output to generate a script for
compiling the perf test object:
--------------------------------------------------
$ cat ~/bin/hello-ebpf
INPUT_FILE=/tmp/test.c
OUTPUT_FILE=/tmp/test.o
export KBUILD_DIR=/lib/modules/5.12.0-0_fbk2_3390_g7ecb4ac46d7f/build
export NR_CPUS=56
export LINUX_VERSION_CODE=0x50c00
export CLANG_EXEC=/data/users/christylee/devtools/llvm/latest/bin/clang
export CLANG_OPTIONS=-xc
export KERNEL_INC_OPTIONS="-nostdinc -isystem
/data/users/christylee/devtools/gcc/10.3.0/lib/gcc/x86_64-pc-linux-gnu/10.3.0/include
-I./arch/\
x86/include -I./arch/x86/include/generated -I./include
-I./arch/x86/include/uapi -I./arch/x86/include/generated/uapi
-I./include/uapi -I./in\
clude/generated/uapi -include ./include/linux/compiler-version.h
-include ./include/linux/kconfig.h"
export PERF_BPF_INC_OPTIONS=-I/home/christylee/lib/perf/include/bpf
export WORKING_DIR=/lib/modules/5.12.0-0_fbk2_3390_g7ecb4ac46d7f/build
export CLANG_SOURCE=-
rm -f $OUTPUT_FILE
cat $INPUT_FILE |
/data/users/christylee/devtools/llvm/latest/bin/clang -D__KERNEL__
-D__NR_CPUS__=56 -DLINUX_VERSION_CODE=0x50c00 -xc -I/ho\
me/christylee/lib/perf/include/bpf -nostdinc -isystem
/data/users/christylee/devtools/gcc/10.3.0/lib/gcc/x86_64-pc-linux-gnu/10.3.0/include
\
-I./arch/x86/include -I./arch/x86/include/generated -I./include
-I./arch/x86/include/uapi -I./arch/x86/include/generated/uapi
-I./include/ua\
pi -I./include/generated/uapi -include
./include/linux/compiler-version.h -include ./include/linux/kconfig.h
-Wno-unused-value -Wno-pointer-\
sign -working-directory
/lib/modules/5.12.0-0_fbk2_3390_g7ecb4ac46d7f/build -c - -target bpf
-O2 -o $OUTPUT_FILE
--------------------------------------------------
I then wrote and compiled a script that ask to get asks to put a probe
at a function that
does not exists in the kernel, it errors out as expected:
$ cat /tmp/test.c
__attribute__((section("fork=does_not_exist"), used)) int fork(void *ctx) {
return 0;
}
char _license[] __attribute__((section("license"), used)) = "GPL";
int _version __attribute__((section("version"), used)) = 0x40100;
$ cd ~/bin && ./hello-ebpf
$ perf record --event /tmp/test.o sleep 1
Using perf wrapper that supports hot-text. Try perf.real if you
encounter any issues.
Probe point 'does_not_exist' not found.
event syntax error: '/tmp/test.o'
\___ You need to check probing points in BPF file
(add -v to see detail)
Run 'perf list' for a list of valid events
Usage: perf record [<options>] [<command>]
or: perf record [<options>] -- <command> [<options>]
-e, --event <event> event selector. use 'perf list' to list
available events
---------------------------------------------------
Next I changed the attribute to something that exists in the kernel.
As expected, it errors out
with permission problem:
$ cat /tmp/test.c
__attribute__((section("fork=fork_init"), used)) int fork(void *ctx) {
return 0;
}
char _license[] __attribute__((section("license"), used)) = "GPL";
int _version __attribute__((section("version"), used)) = 0x40100;
$ grep fork_init /proc/kallsyms
ffffffff8146e250 T xfs_ifork_init_cow
ffffffff83980481 T fork_init
$ cd ~/bin && ./hello-ebpf
$ perf record --event /tmp/test.o sleep 1
Using perf wrapper that supports hot-text. Try perf.real if you
encounter any issues.
Failed to open kprobe_events: Permission denied
event syntax error: '/tmp/test.o'
\___ You need to be root
(add -v to see detail)
Run 'perf list' for a list of valid events
Usage: perf record [<options>] [<command>]
or: perf record [<options>] -- <command> [<options>]
-e, --event <event> event selector. use 'perf list' to list
available events
---------------------------------------------------
So I reran as root, but this time I get an invalid syntax error:
# perf record --event /tmp/test.o -v sleep 1
Using perf wrapper that supports hot-text. Try perf.real if you
encounter any issues.
Failed to write event: Invalid argument
event syntax error: '/tmp/test.o'
\___ Invalid argument
(add -v to see detail)
Run 'perf list' for a list of valid events
Usage: perf record [<options>] [<command>]
or: perf record [<options>] -- <command> [<options>]
-e, --event <event> event selector. use 'perf list' to list
available events
---------------------------------------------------
Is there a different way to attach a custom event probe point?
