Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Thu, 30 Apr 2020 16:08:05 -0500
From:   Josh Poimboeuf <jpoimboe@redhat.com>
To:     Arnd Bergmann <arnd@arndb.de>
Cc:     Peter Zijlstra <peterz@infradead.org>,
        the arch/x86 maintainers <x86@kernel.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Kees Cook <keescook@chromium.org>
Subject: Re: Remaining randconfig objtool warnings, linux-next-20200428
Message-ID: <20200430210805.67h5la6m7rtygeq5@treble>
References: <CAK8P3a2qLJkokMGt48JRky=WUeAbJRuNmoD1oqfWdrGSC6y1LA@mail.gmail.com>
 <20200428161044.caamvx67t2z4t6vd@treble>
 <20200429185536.5xshpcwtn4be4llh@treble>
 <CAK8P3a0M9qh2-_5VKx89ZsTfy5S1zhfWwnO7rN4xYhDwBBvPjw@mail.gmail.com>
 <20200429231115.z2mo5bsmrmj4oark@treble>
 <20200429232835.yw6ajtjyleob3lmz@treble>
 <CAK8P3a07P49-AdHForem=V0_cfiTFSLLhHCEQk6-UDBOg3-Q_g@mail.gmail.com>
 <20200430143350.qezebqmx2xwdxqxq@treble>
 <20200430194630.jkwysx6eftkaf6bu@treble>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <20200430194630.jkwysx6eftkaf6bu@treble>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, Apr 30, 2020 at 02:46:30PM -0500, Josh Poimboeuf wrote:
> > Yup, makes sense.  I think I've been getting confused by the syscall
> > wrappers.
> > 
> > > How hard would it be to change objtool instead of changing the sources?

So I just realized this is an objtool bug after all.  Or at least a new
GCC quirk.  The quick fix is:

  sed -si 's/cold./cold/' tools/objtool/check.c

However... after studying how all this works, I'm actually thinking that
it makes sense to move the SYS_NI usage into kernel/sys_ni.c, like you
originally suggested.  That seems cleaner to me: all the syscall
aliasing code together in one file.  SYS_NI is similar to COND_SYSCALL,
except it has a custom ENOSYS handler.  Having "NI" in the name is
another clue it belongs in sys_ni.c.

Alternatively, I could do the x86-specific SYSCALL_ALIAS, which is easy
enough, but I really prefer the sys_ni.c approach.

Either of those would allow the removal of some hacky objtool code,
which only ever existed in the first place because of posix-stubs.c and
that inline asm SYSCALL_ALIAS macro.

One of objtool's goals is to standardize ELF data, and it seems
reasonable to require the use of C-based aliases.  And I think it would
be a nice cleanup anyway.

-- 
Josh