On Fri, Aug 23, 2024 at 7:48 AM Alexei Starovoitov
<alexei.starovoitov@xxxxxxxxx> wrote:
>
> On Wed, Aug 7, 2024 at 8:17 AM Andrii Nakryiko
> <andrii.nakryiko@xxxxxxxxx> wrote:
> >
> > On Wed, Aug 7, 2024 at 2:07 AM Jiri Olsa <olsajiri@xxxxxxxxx> wrote:
> > >
> > > On Tue, Aug 06, 2024 at 04:03:19PM -0700, Andrii Nakryiko wrote:
> > >
> > > SNIP
> > >
> > > > ssize_t get_uprobe_offset(const void *addr)
> > > > {
> > > > - size_t start, end, base;
> > > > - char buf[256];
> > > > - bool found = false;
> > > > + size_t start, base, end;
> > > > FILE *f;
> > > > + char buf[256];
> > > > + int err, flags;
> > > >
> > > > f = fopen("/proc/self/maps", "r");
> > > > if (!f)
> > > > return -errno;
> > > >
> > > > - while (fscanf(f, "%zx-%zx %s %zx %*[^\n]\n", &start, &end, buf, &base) == 4) {
> > > > - if (buf[2] == 'x' && (uintptr_t)addr >= start && (uintptr_t)addr < end) {
> > > > - found = true;
> > > > - break;
> > > > + /* requested executable VMA only */
> > > > + err = procmap_query(fileno(f), addr, PROCMAP_QUERY_VMA_EXECUTABLE, &start, &base, &flags);
> > > > + if (err == -EOPNOTSUPP) {
> > > > + bool found = false;
> > > > +
> > > > + while (fscanf(f, "%zx-%zx %s %zx %*[^\n]\n", &start, &end, buf, &base) == 4) {
> > > > + if (buf[2] == 'x' && (uintptr_t)addr >= start && (uintptr_t)addr < end) {
> > > > + found = true;
> > > > + break;
> > > > + }
> > > > + }
> > > > + if (!found) {
> > > > + fclose(f);
> > > > + return -ESRCH;
> > > > }
> > > > + } else if (err) {
> > > > + fclose(f);
> > > > + return err;
> > >
> > > I feel like I commented on this before, so feel free to ignore me,
> > > but this seems similar to the code below, could be in one function
> >
> > Do you mean get_rel_offset()? That one is for data symbols (USDT
> > semaphores), so it a) doesn't do arch-specific adjustments and b)
> > doesn't filter by executable flag. So while the logic of parsing and
> > finding VMA is similar, conditions and adjustments are different. It
> > feels not worth combining them, tbh.
> >
> > >
> > > anyway it's good for follow up
> > >
> > > there was another selftest in the original patchset adding benchmark
> > > for the procfs query interface, is it coming in as well?
> >
> > I didn't plan to send it, given it's not really a test. But I can put
> > it on Github somewhere, probably, if it's useful.
>
> With and without this selftest applied I see:
> ./test_progs -t uprobe
> #416 uprobe:OK
> #417 uprobe_autoattach:OK
> [ 47.448908] ref_ctr_offset mismatch. inode: 0x16b5f921 offset:
> 0x2d4297 ref_ctr_offset(old): 0x45e8b56 ref_ctr_offset(new): 0x45e8b54
> #418/1 uprobe_multi_test/skel_api:OK
>
> Is this a known issue?
Yeah, that's not due to my changes. It's an old warning in uprobe
internals, but I think we should remove it, because it can trivially
be triggered by a user. Which is what Jiri is doing intentionally in
one of selftests to test uprobe failure handling.
Jiri, maybe let's get rid of this warning?
>
> Applied anyway.
Thanks! I just found another auto-archived patch of mine, the one
adding multi-uprobe benchmarks (see patchworks). Please take a look
and maybe apply, when you get a chance.
