| Message ID | 34708481d71ea72c23a78a5209e04a76b261a01d.1717881178.git.dxu@dxuuu.xyz |
|---|---|
| State | New |
| Headers | show |
| Series | bpf: Support dumping kfunc prototypes from BTF | expand |
On Sat, Jun 08, 2024 at 03:16:02PM -0600, Daniel Xu wrote: > The prototype defined in bpf_kfuncs.h was not in line with how the > actual kfunc was defined. This causes compilation errors when kfunc > prototypes are generated from BTF. > > Fix by aligning with actual kfunc definition. > > Signed-off-by: Daniel Xu <dxu@dxuuu.xyz> > --- > tools/testing/selftests/bpf/bpf_kfuncs.h | 2 +- > tools/testing/selftests/bpf/progs/kprobe_multi_session_cookie.c | 2 +- > 2 files changed, 2 insertions(+), 2 deletions(-) > > diff --git a/tools/testing/selftests/bpf/bpf_kfuncs.h b/tools/testing/selftests/bpf/bpf_kfuncs.h > index be91a6919315..3b6675ab4086 100644 > --- a/tools/testing/selftests/bpf/bpf_kfuncs.h > +++ b/tools/testing/selftests/bpf/bpf_kfuncs.h > @@ -77,5 +77,5 @@ extern int bpf_verify_pkcs7_signature(struct bpf_dynptr *data_ptr, > struct bpf_key *trusted_keyring) __ksym; > > extern bool bpf_session_is_return(void) __ksym __weak; > -extern long *bpf_session_cookie(void) __ksym __weak; > +extern __u64 *bpf_session_cookie(void) __ksym __weak; the original intent was to expose long instead of __u64 :-\ could we rather change the bpf_session_cookie function to return long? should be just return value type change thanks, jirka > #endif > diff --git a/tools/testing/selftests/bpf/progs/kprobe_multi_session_cookie.c b/tools/testing/selftests/bpf/progs/kprobe_multi_session_cookie.c > index d49070803e22..0835b5edf685 100644 > --- a/tools/testing/selftests/bpf/progs/kprobe_multi_session_cookie.c > +++ b/tools/testing/selftests/bpf/progs/kprobe_multi_session_cookie.c > @@ -25,7 +25,7 @@ int BPF_PROG(trigger) > > static int check_cookie(__u64 val, __u64 *result) > { > - long *cookie; > + __u64 *cookie; > > if (bpf_get_current_pid_tgid() >> 32 != pid) > return 1; > -- > 2.44.0 >
Hi Jiri, On Mon, Jun 10, 2024 at 03:04:25PM GMT, Jiri Olsa wrote: > On Sat, Jun 08, 2024 at 03:16:02PM -0600, Daniel Xu wrote: > > The prototype defined in bpf_kfuncs.h was not in line with how the > > actual kfunc was defined. This causes compilation errors when kfunc > > prototypes are generated from BTF. > > > > Fix by aligning with actual kfunc definition. > > > > Signed-off-by: Daniel Xu <dxu@dxuuu.xyz> > > --- > > tools/testing/selftests/bpf/bpf_kfuncs.h | 2 +- > > tools/testing/selftests/bpf/progs/kprobe_multi_session_cookie.c | 2 +- > > 2 files changed, 2 insertions(+), 2 deletions(-) > > > > diff --git a/tools/testing/selftests/bpf/bpf_kfuncs.h b/tools/testing/selftests/bpf/bpf_kfuncs.h > > index be91a6919315..3b6675ab4086 100644 > > --- a/tools/testing/selftests/bpf/bpf_kfuncs.h > > +++ b/tools/testing/selftests/bpf/bpf_kfuncs.h > > @@ -77,5 +77,5 @@ extern int bpf_verify_pkcs7_signature(struct bpf_dynptr *data_ptr, > > struct bpf_key *trusted_keyring) __ksym; > > > > extern bool bpf_session_is_return(void) __ksym __weak; > > -extern long *bpf_session_cookie(void) __ksym __weak; > > +extern __u64 *bpf_session_cookie(void) __ksym __weak; > > the original intent was to expose long instead of __u64 :-\ > > could we rather change the bpf_session_cookie function to return long? > should be just return value type change Sounds reasonable to me. I don't think the kfunc has made it to a release yet, so perhaps if we extract this commit out as a fix to bpf tree it can still make it into 6.10. That way we won't have to worry about any ABI changes. Thanks, Daniel
On Tue, Jun 11, 2024 at 8:54 AM Daniel Xu <dxu@dxuuu.xyz> wrote: > > Hi Jiri, > > On Mon, Jun 10, 2024 at 03:04:25PM GMT, Jiri Olsa wrote: > > On Sat, Jun 08, 2024 at 03:16:02PM -0600, Daniel Xu wrote: > > > The prototype defined in bpf_kfuncs.h was not in line with how the > > > actual kfunc was defined. This causes compilation errors when kfunc > > > prototypes are generated from BTF. > > > > > > Fix by aligning with actual kfunc definition. > > > > > > Signed-off-by: Daniel Xu <dxu@dxuuu.xyz> > > > --- > > > tools/testing/selftests/bpf/bpf_kfuncs.h | 2 +- > > > tools/testing/selftests/bpf/progs/kprobe_multi_session_cookie.c | 2 +- > > > 2 files changed, 2 insertions(+), 2 deletions(-) > > > > > > diff --git a/tools/testing/selftests/bpf/bpf_kfuncs.h b/tools/testing/selftests/bpf/bpf_kfuncs.h > > > index be91a6919315..3b6675ab4086 100644 > > > --- a/tools/testing/selftests/bpf/bpf_kfuncs.h > > > +++ b/tools/testing/selftests/bpf/bpf_kfuncs.h > > > @@ -77,5 +77,5 @@ extern int bpf_verify_pkcs7_signature(struct bpf_dynptr *data_ptr, > > > struct bpf_key *trusted_keyring) __ksym; > > > > > > extern bool bpf_session_is_return(void) __ksym __weak; > > > -extern long *bpf_session_cookie(void) __ksym __weak; > > > +extern __u64 *bpf_session_cookie(void) __ksym __weak; > > > > the original intent was to expose long instead of __u64 :-\ > > > > could we rather change the bpf_session_cookie function to return long? > > should be just return value type change > > Sounds reasonable to me. I don't think the kfunc has made it to a > release yet, so perhaps if we extract this commit out as a fix to bpf > tree it can still make it into 6.10. That way we won't have to worry > about any ABI changes. kfunc-s can be changed at any time. Keep targeting bpf-next for everything.
On Mon, Jun 10, 2024 at 6:04 AM Jiri Olsa <olsajiri@gmail.com> wrote: > > On Sat, Jun 08, 2024 at 03:16:02PM -0600, Daniel Xu wrote: > > The prototype defined in bpf_kfuncs.h was not in line with how the > > actual kfunc was defined. This causes compilation errors when kfunc > > prototypes are generated from BTF. > > > > Fix by aligning with actual kfunc definition. > > > > Signed-off-by: Daniel Xu <dxu@dxuuu.xyz> > > --- > > tools/testing/selftests/bpf/bpf_kfuncs.h | 2 +- > > tools/testing/selftests/bpf/progs/kprobe_multi_session_cookie.c | 2 +- > > 2 files changed, 2 insertions(+), 2 deletions(-) > > > > diff --git a/tools/testing/selftests/bpf/bpf_kfuncs.h b/tools/testing/selftests/bpf/bpf_kfuncs.h > > index be91a6919315..3b6675ab4086 100644 > > --- a/tools/testing/selftests/bpf/bpf_kfuncs.h > > +++ b/tools/testing/selftests/bpf/bpf_kfuncs.h > > @@ -77,5 +77,5 @@ extern int bpf_verify_pkcs7_signature(struct bpf_dynptr *data_ptr, > > struct bpf_key *trusted_keyring) __ksym; > > > > extern bool bpf_session_is_return(void) __ksym __weak; > > -extern long *bpf_session_cookie(void) __ksym __weak; > > +extern __u64 *bpf_session_cookie(void) __ksym __weak; > > the original intent was to expose long instead of __u64 :-\ > Cookies internally are always u64 (8 byte values). Marking them internally in the kernel as long could lead to problems on 32-bit architectures, potentially (it still needs to be 64-bit value according to BPF contract, but we'll allocate only 4 bytes for them). It seems better and safer to be explicit with __u64/u64 for cookies everywhere. What am I missing? > could we rather change the bpf_session_cookie function to return long? > should be just return value type change > > thanks, > jirka > > > > #endif > > diff --git a/tools/testing/selftests/bpf/progs/kprobe_multi_session_cookie.c b/tools/testing/selftests/bpf/progs/kprobe_multi_session_cookie.c > > index d49070803e22..0835b5edf685 100644 > > --- a/tools/testing/selftests/bpf/progs/kprobe_multi_session_cookie.c > > +++ b/tools/testing/selftests/bpf/progs/kprobe_multi_session_cookie.c > > @@ -25,7 +25,7 @@ int BPF_PROG(trigger) > > > > static int check_cookie(__u64 val, __u64 *result) > > { > > - long *cookie; > > + __u64 *cookie; > > > > if (bpf_get_current_pid_tgid() >> 32 != pid) > > return 1; > > -- > > 2.44.0 > >
diff --git a/tools/testing/selftests/bpf/bpf_kfuncs.h b/tools/testing/selftests/bpf/bpf_kfuncs.h index be91a6919315..3b6675ab4086 100644 --- a/tools/testing/selftests/bpf/bpf_kfuncs.h +++ b/tools/testing/selftests/bpf/bpf_kfuncs.h @@ -77,5 +77,5 @@ extern int bpf_verify_pkcs7_signature(struct bpf_dynptr *data_ptr, struct bpf_key *trusted_keyring) __ksym; extern bool bpf_session_is_return(void) __ksym __weak; -extern long *bpf_session_cookie(void) __ksym __weak; +extern __u64 *bpf_session_cookie(void) __ksym __weak; #endif diff --git a/tools/testing/selftests/bpf/progs/kprobe_multi_session_cookie.c b/tools/testing/selftests/bpf/progs/kprobe_multi_session_cookie.c index d49070803e22..0835b5edf685 100644 --- a/tools/testing/selftests/bpf/progs/kprobe_multi_session_cookie.c +++ b/tools/testing/selftests/bpf/progs/kprobe_multi_session_cookie.c @@ -25,7 +25,7 @@ int BPF_PROG(trigger) static int check_cookie(__u64 val, __u64 *result) { - long *cookie; + __u64 *cookie; if (bpf_get_current_pid_tgid() >> 32 != pid) return 1;
The prototype defined in bpf_kfuncs.h was not in line with how the actual kfunc was defined. This causes compilation errors when kfunc prototypes are generated from BTF. Fix by aligning with actual kfunc definition. Signed-off-by: Daniel Xu <dxu@dxuuu.xyz> --- tools/testing/selftests/bpf/bpf_kfuncs.h | 2 +- tools/testing/selftests/bpf/progs/kprobe_multi_session_cookie.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)