Re: [PATCH v2] perf,x86: avoid missing caller address in stack traces captured in uprobe

[Andrii Nakryiko <andrii.nakryiko@gmail.com>]

On Tue, Jul 2, 2024 at 4:39 PM Josh Poimboeuf <jpoimboe@kernel.org> wrote:
>
> On Tue, Jul 02, 2024 at 04:35:56PM -0700, Josh Poimboeuf wrote:
> > On Tue, Jul 02, 2024 at 10:18:58AM -0700, Andrii Nakryiko wrote:
> > > When tracing user functions with uprobe functionality, it's common to
> > > install the probe (e.g., a BPF program) at the first instruction of the
> > > function. This is often going to be `push %rbp` instruction in function
> > > preamble, which means that within that function frame pointer hasn't
> > > been established yet. This leads to consistently missing an actual
> > > caller of the traced function, because perf_callchain_user() only
> > > records current IP (capturing traced function) and then following frame
> > > pointer chain (which would be caller's frame, containing the address of
> > > caller's caller).
> > >
> > > So when we have target_1 -> target_2 -> target_3 call chain and we are
> > > tracing an entry to target_3, captured stack trace will report
> > > target_1 -> target_3 call chain, which is wrong and confusing.
> > >
> > > This patch proposes a x86-64-specific heuristic to detect `push %rbp`
> > > (`push %ebp` on 32-bit architecture) instruction being traced. Given
> > > entire kernel implementation of user space stack trace capturing works
> > > under assumption that user space code was compiled with frame pointer
> > > register (%rbp/%ebp) preservation, it seems pretty reasonable to use
> > > this instruction as a strong indicator that this is the entry to the
> > > function. In that case, return address is still pointed to by %rsp/%esp,
> > > so we fetch it and add to stack trace before proceeding to unwind the
> > > rest using frame pointer-based logic.
> > >
> > > Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
> >
> > Should it also check for ENDBR64?
> >

Sure, I can add a check for endbr64 as well. endbr64 probably can be
used not just at function entry, is that right? So it might be another
case of false positive (which I think is ok, see below).

> > When compiled with -fcf-protection=branch, the first instruction of the
> > function will almost always be ENDBR64.  I'm not sure about other
> > distros, but at least Fedora compiles its binaries like that.
>
> BTW, there are some cases (including leaf functions and some stack
> alignment sequences) where a "push %rbp" can happen inside a function.
> Then it would presumably add a bogus trace entry.  Are such false
> positives ok?

I think such cases should be rare. People mostly seem to trace user
function entry/exit, rarely if ever they trace something within the
function, except for USDT cases, where it will be a nop instruction
that they trace.

In general, even with false positives, I think it's overwhelmingly
better to get correct entry stack trace 99.9% of the time, and in the
rest 0.01% cases it's fine having one extra bogus entry (but the rest
should still be correct), which should be easy for humans to recognize
and filter out, if necessary.

>
> --
> Josh