netdev - Re: [PATCH net-next 1/2] bpf: improve htab inlining for future 32 bit jits

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <43a2b95f-db1d-8c33-a7d6-42f8299ddcb3@fb.com>
Date:   Fri, 18 Aug 2017 17:00:57 -0700
From:   Alexei Starovoitov <ast@...com>
To:     Daniel Borkmann <daniel@...earbox.net>, <davem@...emloft.net>
CC:     <netdev@...r.kernel.org>
Subject: Re: [PATCH net-next 1/2] bpf: improve htab inlining for future 32 bit
 jits

On 8/18/17 4:51 PM, Daniel Borkmann wrote:
> Lets future proof htab lookup inlining, commit 9015d2f59535 ("bpf:
> inline htab_map_lookup_elem()") was making the assumption that a
> direct call emission to __htab_map_lookup_elem() will always work
> out for JITs. This is currently true since all JITs we have are
> for 64 bit archs, but in case of 32 bit JITs like upcoming arm32,
> we get a NULL pointer dereference when executing the call to
> __htab_map_lookup_elem() since passed arguments are of a different
> size (unsigned long vs. u64 for pointers) than what we do out of
> BPF. Thus, lets do a proper BPF_CALL_2() declaration such that we
> don't need to make any such assumptions.
>
> Reported-by: Shubham Bansal <illusionist.neo@...il.com>
> Signed-off-by: Daniel Borkmann <daniel@...earbox.net>

assuming on 64-bit archs the should be no perf difference
and only increase in .text, since __htab_map_lookup_elem
is now force inlined into a bunch of places?
I guess that's ok, but kinda sux for 64-bit archs to pay
such penalty because of 32-bit archs.
May be drop always_inline and do such thing conditionally
on 32-bit archs only?
what's the increase in .text?
any difference seen in map_perf_test?