On Fri, Mar 24, 2023 at 8:09 PM Stanislav Fomichev <sdf@xxxxxxxxxx> wrote: > > On 03/24, JP Kobryn wrote: > > This patch prevents races on the print function pointer, allowing the > > libbpf_set_print() function to become thread safe. > > Why does it have to be thread-safe? The rest of the APIs aren't, so It doesn't have to, but if it can be made thread-safe trivially, it probably should be. Rust users are especially sensitive to this (see discussion in [0] for example). Generally speaking, yep, libbpf APIs are not thread-safe by default (we don't do explicit locking anywhere inside libbpf), but there are a bunch of APIs that are inherently thread-safe as they are stateless (like feature probing, string conversion APIs, I suspect upon inspection we'll see that all bpf_program__attach_*() APIs are also probably thread-safe already). So we should start marking them as such to avoid confusion and uncertainty for users. I'd also like to document somewhere that two independent bpf_objects can be used safely on two separate threads, because whatever state they are sharing (like feature detection cache) is designed in such a way to be thread-safe and shareable with no locking. [0] https://github.com/libbpf/libbpf-rs/pull/374#issuecomment-1462565778 > why can't use solve it on your side by wrapping those calls with a > mutex? It would be very unfortunate to wrap libbpf_set_print and *all other libbpf API* in mutex. > > (is there some context I'm missing?) > > > Signed-off-by: JP Kobryn <inwardvessel@xxxxxxxxx> > > --- > > tools/lib/bpf/libbpf.c | 9 ++++++--- > > tools/lib/bpf/libbpf.h | 3 +++ > > 2 files changed, 9 insertions(+), 3 deletions(-) > > > diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c > > index f6a071db5c6e..15737d7b5a28 100644 > > --- a/tools/lib/bpf/libbpf.c > > +++ b/tools/lib/bpf/libbpf.c > > @@ -216,9 +216,10 @@ static libbpf_print_fn_t __libbpf_pr = __base_pr; > > > libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn) > > { > > - libbpf_print_fn_t old_print_fn = __libbpf_pr; > > + libbpf_print_fn_t old_print_fn; > > + > > + old_print_fn = __atomic_exchange_n(&__libbpf_pr, fn, __ATOMIC_RELAXED); > > > - __libbpf_pr = fn; > > return old_print_fn; > > } > > > @@ -227,8 +228,10 @@ void libbpf_print(enum libbpf_print_level level, > > const char *format, ...) > > { > > va_list args; > > int old_errno; > > + libbpf_print_fn_t print_fn; > > > - if (!__libbpf_pr) > > + print_fn = __atomic_load_n(&__libbpf_pr, __ATOMIC_RELAXED); > > + if (!print_fn) > > return; > > > old_errno = errno; > > diff --git a/tools/lib/bpf/libbpf.h b/tools/lib/bpf/libbpf.h > > index 1615e55e2e79..4478809ff9ca 100644 > > --- a/tools/lib/bpf/libbpf.h > > +++ b/tools/lib/bpf/libbpf.h > > @@ -99,6 +99,9 @@ typedef int (*libbpf_print_fn_t)(enum > > libbpf_print_level level, > > /** > > * @brief **libbpf_set_print()** sets user-provided log callback > > function to > > * be used for libbpf warnings and informational messages. > > + * > > + * This function is thread safe. > > + * > > * @param fn The log print function. If NULL, libbpf won't print > > anything. > > * @return Pointer to old print function. > > */ > > -- > > 2.39.2 >
