Received-SPF: pass (google.com: domain of linux-kernel+bounces-394-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:4601:e00::3 as permitted sender) client-ip=2604:1380:4601:e00::3;
Precedence: bulk
MIME-Version: 1.0
References: <cover.1702594357.git.dxu@dxuuu.xyz> <97ada80f3aaaeb16bf97e31a8fc204513b4fb6a9.1702594357.git.dxu@dxuuu.xyz>
 <CAADnVQ+j80DCDHsqJZVmBOohFzOT01Ofdi3TbFEuV2xJ4+A=tA@mail.gmail.com>
In-Reply-To: <CAADnVQ+j80DCDHsqJZVmBOohFzOT01Ofdi3TbFEuV2xJ4+A=tA@mail.gmail.com>
From: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Date: Thu, 14 Dec 2023 19:10:58 -0800
Message-ID: <CAADnVQLtQ3Qcv3Fp9iNVCXY-_2zTP9hEFqccKLxjgUZNs_uWjw@mail.gmail.com>
Subject: Re: [PATCH bpf-next 2/3] bpf: selftests: Add bpf_assert_if() and
 bpf_assert_with_if() macros
To: Daniel Xu <dxu@dxuuu.xyz>
Cc: Alexei Starovoitov <ast@kernel.org>, Daniel Borkmann <daniel@iogearbox.net>, 
	Andrii Nakryiko <andrii@kernel.org>, Shuah Khan <shuah@kernel.org>, 
	Kumar Kartikeya Dwivedi <memxor@gmail.com>, Martin KaFai Lau <martin.lau@linux.dev>, Song Liu <song@kernel.org>, 
	Yonghong Song <yonghong.song@linux.dev>, John Fastabend <john.fastabend@gmail.com>, 
	KP Singh <kpsingh@kernel.org>, Stanislav Fomichev <sdf@google.com>, Hao Luo <haoluo@google.com>, 
	Jiri Olsa <jolsa@kernel.org>, Mykola Lysenko <mykolal@fb.com>, bpf <bpf@vger.kernel.org>, 
	"open list:KERNEL SELFTEST FRAMEWORK" <linux-kselftest@vger.kernel.org>, LKML <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Thu, Dec 14, 2023 at 6:46=E2=80=AFPM Alexei Starovoitov
<alexei.starovoitov@gmail.com> wrote:
>
> On Thu, Dec 14, 2023 at 2:56=E2=80=AFPM Daniel Xu <dxu@dxuuu.xyz> wrote:
> >
> > These macros are a temporary stop-gap until bpf exceptions support
> > unwinding acquired entities. Basically these macros act as if they take
> > a callback which only get executed if the assertion fails.
> >
> > Signed-off-by: Daniel Xu <dxu@dxuuu.xyz>
> > ---
> >  .../testing/selftests/bpf/bpf_experimental.h  | 22 +++++++++++++++++++
> >  1 file changed, 22 insertions(+)
> >
> > diff --git a/tools/testing/selftests/bpf/bpf_experimental.h b/tools/tes=
ting/selftests/bpf/bpf_experimental.h
> > index 1386baf9ae4a..d63f415bef26 100644
> > --- a/tools/testing/selftests/bpf/bpf_experimental.h
> > +++ b/tools/testing/selftests/bpf/bpf_experimental.h
> > @@ -263,6 +263,17 @@ extern void bpf_throw(u64 cookie) __ksym;
> >   */
> >  #define bpf_assert(cond) if (!(cond)) bpf_throw(0);
> >
> > +/* Description
> > + *     Assert that a conditional expression is true. If false, runs co=
de in the
> > + *     body before throwing.
> > + * Returns
> > + *     Void.
> > + * Throws
> > + *     An exception with the value zero when the assertion fails.
> > + */
> > +#define bpf_assert_if(cond) \
> > +       for (int ___i =3D 0, ___j =3D !!(cond); !(___j) && !___i; bpf_t=
hrow(0), ___i++)
>
> Kumar,
>
> Is this approach reliable?
> I suspect the compiler can still optimize it.
> I feel it will be annoying to clean up if folks start using it now,
> since there won't be a drop in replacement.
> Every such bpf_assert_if() would need to be manually patched.
>
> If 2nd part of exception is far, how about we add an equivalent
> of __bpf_assert() macroses with conditional ops in asm,
> but with extra 'asm volatile goto' that can be used to construct
> release of resources.
>
> bpf_do_assert_eq(var1, 0) { bpf_spin_unlock(...); }
> bpf_do_assert_lt(var2, 0) { bpf_spin_unlock(...); }

Just realized that we can go the other way instead.

We can get rid of bpf_assert_eq/ne/... and replace with:

diff --git a/tools/testing/selftests/bpf/bpf_experimental.h
b/tools/testing/selftests/bpf/bpf_experimental.h
index 1386baf9ae4a..1c500287766d 100644
--- a/tools/testing/selftests/bpf/bpf_experimental.h
+++ b/tools/testing/selftests/bpf/bpf_experimental.h
@@ -254,6 +254,15 @@ extern void bpf_throw(u64 cookie) __ksym;
                }
                 \
         })

+#define _EQ(LHS, RHS) \
+       ({ int var =3D 1;\
+               asm volatile goto("if %[lhs] =3D=3D %[rhs] goto %l[l_yes]" =
\
+               :: [lhs] "r"(LHS), [rhs] "i"(RHS) :: l_yes);\
+       var =3D 0;\
+l_yes:\
+       var;\
+       })
+
 /* Description
  *     Assert that a conditional expression is true.
  * Returns
diff --git a/tools/testing/selftests/bpf/progs/exceptions.c
b/tools/testing/selftests/bpf/progs/exceptions.c
index 2811ee842b01..1111e852f154 100644
--- a/tools/testing/selftests/bpf/progs/exceptions.c
+++ b/tools/testing/selftests/bpf/progs/exceptions.c
@@ -203,6 +203,7 @@ __noinline int assert_nz_gfunc(u64 c)
        volatile u64 cookie =3D c;

        bpf_assert(cookie !=3D 0);
+       bpf_assert(_EQ(cookie, 2));
        return 0;
 }

we can probably remove bpf_assert_with() and
all of the bpf_assert_le|ne|qt|eq|_with()

Users can open code everything:
if (!_EQ(foo, bar)) bpf_throw(123);

bpf_assert_if() can work too,
but let's call it bpf_do_assert() and use like:

bpf_do_assert(EQ(time, 0)) {
   // cleanup
}