MIME-Version: 1.0
References: <20221101222520.never.109-kees@kernel.org> <20221101223321.1326815-5-keescook@chromium.org>
 <CAMSo37W3gRkP02tSCxGX71ZDAt3WgPZrkTRTM6J1iQ4gvUS9vg@mail.gmail.com>
 <CANDhNCogJrvt=yEXFK-xVmGjkcRxSNGZUqUeNw2MV9bFRrwPdQ@mail.gmail.com>
 <63dbf04a.630a0220.2608a.0149@mx.google.com> <CANDhNCoJJs_jSAdr6uved=cK=-6+nzSJq3e_E3HgtiXHZXkVww@mail.gmail.com>
 <63dc0a89.170a0220.49507.0091@mx.google.com> <CAMj1kXHTwFH-OADZ5nZXfZyuH-x7XJL3DMyiz2Tr_nCuup+xFA@mail.gmail.com>
 <CAKwvOd=cuDb2NNK4uoUTYTLryFig47hvDQ08ta5XhjHsEhAG4A@mail.gmail.com>
In-Reply-To: <CAKwvOd=cuDb2NNK4uoUTYTLryFig47hvDQ08ta5XhjHsEhAG4A@mail.gmail.com>
From:   Sami Tolvanen <samitolvanen@google.com>
Date:   Thu, 2 Feb 2023 11:49:42 -0800
Message-ID: <CABCJKuc+GvTHXu-CVedZs6CvPKJ8ySrVztkP2V3dsWn70xf7hA@mail.gmail.com>
Subject: Re: [PATCH 5/6] driver core: Add __alloc_size hint to devm allocators
To:     Nick Desaulniers <ndesaulniers@google.com>
Cc:     Ard Biesheuvel <ardb@kernel.org>,
        Kees Cook <keescook@chromium.org>,
        John Stultz <jstultz@google.com>,
        Yongqin Liu <yongqin.liu@linaro.org>,
        Vlastimil Babka <vbabka@suse.cz>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Rasmus Villemoes <rasmus.villemoes@prevas.dk>,
        Thomas Gleixner <tglx@linutronix.de>,
        Jason Gunthorpe <jgg@ziepe.ca>, Nishanth Menon <nm@ti.com>,
        Michael Kelley <mikelley@microsoft.com>,
        Dan Williams <dan.j.williams@intel.com>,
        Won Chung <wonchung@google.com>,
        David Gow <davidgow@google.com>,
        Christoph Lameter <cl@linux.com>,
        Pekka Enberg <penberg@kernel.org>,
        David Rientjes <rientjes@google.com>,
        Joonsoo Kim <iamjoonsoo.kim@lge.com>,
        Roman Gushchin <roman.gushchin@linux.dev>,
        Hyeonggon Yoo <42.hyeyoo@gmail.com>,
        Guenter Roeck <linux@roeck-us.net>,
        Andy Shevchenko <andriy.shevchenko@intel.com>,
        Paolo Abeni <pabeni@redhat.com>,
        Geert Uytterhoeven <geert@linux-m68k.org>,
        Nathan Chancellor <nathan@kernel.org>,
        Tom Rix <trix@redhat.com>, linux-kernel@vger.kernel.org,
        linux-mm@kvack.org, linux-hardening@vger.kernel.org,
        llvm@lists.linux.dev, Sumit Semwal <sumit.semwal@linaro.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Thu, Feb 2, 2023 at 11:31 AM Nick Desaulniers
<ndesaulniers@google.com> wrote:
>
> On Thu, Feb 2, 2023 at 11:20 AM Ard Biesheuvel <ardb@kernel.org> wrote:
> >
> > On Thu, 2 Feb 2023 at 20:10, Kees Cook <keescook@chromium.org> wrote:
> > >
> > > On Thu, Feb 02, 2023 at 10:56:29AM -0800, John Stultz wrote:
> > > > That said, making sense of the error message isn't completely trivial
> > > > either. I've been seeing a few cases recently of some of the new
> > > > compiler tooling (I pinged you earlier on a CFI one) causing errors
> > > > that developers aren't really sure how to address.  I know sometimes
> > > > it's not easy to surface the errors with context to what was wrong,
> > > > but at the risk of intense bike shedding, is there some way to provide
> > > > something like "Likely array bounds error" instead of just "BRK
> > > > handler: Fatal exception"?
> > >
> > > Yeah, this is a result of the size trade-off that resulted in config
> > > CONFIG_UBSAN_TRAP -- there ends up being no message about what went
> > > wrong. I'd really like to have cleaner handling of this -- perhaps what
> > > was done for KCFI could be applied to UBSAN as well, though this is an
> > > area I don't know well myself. (i.e. encoding "this was a UBSAN trap"
> > > in the trap itself.)
> > >
> > > Sami or Ard, is this something that could be improved for arm64?
> > >
> >
> > -ENOCONTEXT, so I am going to assume this is about runtime
> > instrumentation that needs some kind of 'panic' function which it will
> > invoke if some condition is met that should never occur?
> >
> > We already use brk with different immediate values in the opcode, so
> > the arch layer already has what we need. Is this a limitation in the
> > compiler, perhaps, where it always emits the same brk opcode?
>
> Yeah, we'd need to update both the compiler to produce the encoding,
> and the kernel to recognize the encoding and do something special.

A quick look at Clang's source code suggests that Intrinsic::ubsantrap
already accepts the handler ID (from the SanitizerHandler enum) as an
argument and the arm64 LLVM back-end appears to encode the value as an
immediate for the brk instruction. I didn't confirm that this actually
works, but perhaps we just need to teach the kernel about the possible
values?

Sami