Received-SPF: pass (google.com: domain of linux-kernel+bounces-62170-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
Precedence: bulk
MIME-Version: 1.0
References: <20240209-hid-bpf-sleepable-v1-0-4cc895b5adbd@kernel.org>
 <87bk8pve2z.fsf@toke.dk> <CAO-hwJ+UeaBydN9deA8KBbgBiC_UCt6oXX-wGnNuSr8fhUrkXw@mail.gmail.com>
 <875xyxva9u.fsf@toke.dk> <CAO-hwJLvEGNRXc8G2PR+AQ6kJg+k5YqSt3F7LCSc0zWnmFfe5g@mail.gmail.com>
 <87r0hhfudh.fsf@toke.dk>
In-Reply-To: <87r0hhfudh.fsf@toke.dk>
From: Benjamin Tissoires <benjamin.tissoires@redhat.com>
Date: Mon, 12 Feb 2024 19:20:53 +0100
Message-ID: <CAO-hwJLxkt=THKBjxDA6KZsC5h52rCXZ-2RNKPCiYMHNjhQJNg@mail.gmail.com>
Subject: Re: [PATCH RFC bpf-next 0/9] allow HID-BPF to do device IOs
To: =?UTF-8?B?VG9rZSBIw7hpbGFuZC1Kw7hyZ2Vuc2Vu?= <toke@redhat.com>
Cc: Benjamin Tissoires <bentiss@kernel.org>, Alexei Starovoitov <ast@kernel.org>, 
	Daniel Borkmann <daniel@iogearbox.net>, John Fastabend <john.fastabend@gmail.com>, 
	Andrii Nakryiko <andrii@kernel.org>, Martin KaFai Lau <martin.lau@linux.dev>, 
	Eduard Zingerman <eddyz87@gmail.com>, Song Liu <song@kernel.org>, 
	Yonghong Song <yonghong.song@linux.dev>, KP Singh <kpsingh@kernel.org>, 
	Stanislav Fomichev <sdf@google.com>, Hao Luo <haoluo@google.com>, Jiri Olsa <jolsa@kernel.org>, 
	Jiri Kosina <jikos@kernel.org>, Jonathan Corbet <corbet@lwn.net>, Shuah Khan <shuah@kernel.org>, 
	bpf@vger.kernel.org, linux-kernel@vger.kernel.org, 
	linux-input@vger.kernel.org, linux-doc@vger.kernel.org, 
	linux-kselftest@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Mon, Feb 12, 2024 at 6:46=E2=80=AFPM Toke H=C3=B8iland-J=C3=B8rgensen <t=
oke@redhat.com> wrote:
>
> Benjamin Tissoires <benjamin.tissoires@redhat.com> writes:
>
> [...]
> >> IIUC, the bpf_timer callback is just a function (subprog) from the
> >> verifier PoV, so it is verified as whatever program type is creating t=
he
> >> timer. So in other words, as long as you setup the timer from inside a
> >> tracing prog type, you should have access to all the same kfuncs, I
> >> think?
> >
> > Yep, you are correct. But as mentioned above, I am now in trouble with
> > the sleepable state:
> > - I need to call timer_start() from a non sleepable tracing function
> > (I'm in hard IRQ when dealing with a physical device)
> > - but then, ideally, the callback function needs to be tagged as a
> > sleepable one, so I can export my kfuncs which are doing kzalloc and
> > device IO as such.
> >
> > However, I can not really teach the BPF verifier to do so:
> > - it seems to check for the callback first when it is loaded, and
> > there is no SEC() equivalent for static functions
> > - libbpf doesn't have access to the callback as a prog as it has to be
> > a static function, and thus isn't exported as a full-blown prog.
> > - the verifier only checks for the callback when dealing with
> > BPF_FUNC_timer_set_callback, which doesn't have a "flag" argument
> > (though the validation of the callback has already been done while
> > checking it first, so we are already too late to change the sleppable
> > state of the callback)
> >
> > Right now, the only OK-ish version I have is declaring the kfunc as
> > non-sleepable, but checking that we are in a different context than
> > the IRQ of the initial event. This way, I am not crashing if this
> > function is called from the initial IRQ, but will still crash if used
> > outside of the hid context.
> >
> > This is not satisfactory, but I feel like it's going to be hard to
> > teach the verifier that the callback function is sleepable in that
> > case (maybe we could suffix the callback name, like we do for
> > arguments, but this is not very clean either).
>
> The callback is only set once when the timer is first setup; I *think*
> it works to do the setup (bpf_timer_init() and bpf_timer_set_callback())
> in the context you need (from a sleepable prog), but do the arming
> (bpf_timer_start()) from a different program that is not itself sleepable=
?
>

Genius! It works, and I can just keep having them declared as a
syscall kfunc, not as a tracing kfunc.

But isn't this an issue outside of my use case? I mean, if the
callback is assuming the environment for when it is set up but can be
called from any context there seems to be a problem when 2 contexts
are not equivalent, no?