Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:18 as permitted sender) client-ip=2620:137:e000::1:18;
From:   Jiri Olsa <olsajiri@gmail.com>
Date:   Wed, 18 May 2022 13:24:56 +0200
To:     Yonghong Song <yhs@fb.com>
Cc:     Jiri Olsa <olsajiri@gmail.com>,
        Eugene Syromiatnikov <esyr@redhat.com>,
        Masami Hiramatsu <mhiramat@kernel.org>,
        Steven Rostedt <rostedt@goodmis.org>,
        Ingo Molnar <mingo@redhat.com>,
        Alexei Starovoitov <ast@kernel.org>,
        Daniel Borkmann <daniel@iogearbox.net>,
        Andrii Nakryiko <andrii@kernel.org>,
        Martin KaFai Lau <kafai@fb.com>,
        Song Liu <songliubraving@fb.com>,
        John Fastabend <john.fastabend@gmail.com>,
        KP Singh <kpsingh@kernel.org>, netdev@vger.kernel.org,
        bpf@vger.kernel.org, linux-kernel@vger.kernel.org,
        Shuah Khan <shuah@kernel.org>, linux-kselftest@vger.kernel.org
Subject: Re: [PATCH bpf-next v3 4/4] bpf_trace: pass array of u64 values in
 kprobe_multi.addrs
Message-ID: <YoTXiAk1EpZ0rLKE@krava>
References: <cover.1652772731.git.esyr@redhat.com>
 <6ef675aeeea442fa8fc168cd1cb4e4e474f65a3f.1652772731.git.esyr@redhat.com>
 <YoNnAgDsIWef82is@krava>
 <20220517123050.GA25149@asgard.redhat.com>
 <YoP/eEMqAn3sVFXf@krava>
 <7c5e64f2-f2cf-61b7-9231-fc267bf0f2d8@fb.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <7c5e64f2-f2cf-61b7-9231-fc267bf0f2d8@fb.com>
Precedence: bulk

On Tue, May 17, 2022 at 02:34:55PM -0700, Yonghong Song wrote:
> 
> 
> On 5/17/22 1:03 PM, Jiri Olsa wrote:
> > On Tue, May 17, 2022 at 02:30:50PM +0200, Eugene Syromiatnikov wrote:
> > > On Tue, May 17, 2022 at 11:12:34AM +0200, Jiri Olsa wrote:
> > > > On Tue, May 17, 2022 at 09:36:47AM +0200, Eugene Syromiatnikov wrote:
> > > > > With the interface as defined, it is impossible to pass 64-bit kernel
> > > > > addresses from a 32-bit userspace process in BPF_LINK_TYPE_KPROBE_MULTI,
> > > > > which severly limits the useability of the interface, change the ABI
> > > > > to accept an array of u64 values instead of (kernel? user?) longs.
> > > > > Interestingly, the rest of the libbpf infrastructure uses 64-bit values
> > > > > for kallsyms addresses already, so this patch also eliminates
> > > > > the sym_addr cast in tools/lib/bpf/libbpf.c:resolve_kprobe_multi_cb().
> > > > 
> > > > so the problem is when we have 32bit user sace on 64bit kernel right?
> > > > 
> > > > I think we should keep addrs as longs in uapi and have kernel to figure out
> > > > if it needs to read u32 or u64, like you did for symbols in previous patch
> > > 
> > > No, it's not possible here, as addrs are kernel addrs and not user space
> > > addrs, so user space has to explicitly pass 64-bit addresses on 64-bit
> > > kernels (or have a notion whether it is running on a 64-bit
> > > or 32-bit kernel, and form the passed array accordingly, which is against
> > > the idea of compat layer that tries to abstract it out).
> > 
> > hum :-\ I'll need to check on compat layer.. there must
> > be some other code doing this already somewhere, right?

so the 32bit application running on 64bit kernel using libbpf won't
work at the moment, right? because it sees:

  bpf_kprobe_multi_opts::addrs as its 'unsigned long'

which is 4 bytes and it needs to put there 64bits kernel addresses

if we force the libbpf interface to use u64, then we should be fine

> 
> I am not familiar with all these compatibility thing. But if we
> have 64-bit pointer for **syms, maybe we could also have
> 64-bit pointer for *syms for consistency?

right, perhaps we could have one function to read both syms and addrs arrays

> 
> > jirka
> > 
> > > 
> > > > we'll need to fix also bpf_kprobe_multi_cookie_swap because it assumes
> > > > 64bit user space pointers

if we have both addresses and cookies 64 then this should be ok

> > > > 
> > > > would be gret if we could have selftest for this

let's add selftest for this

thanks,
jirka