MIME-Version: 1.0
References: <CA+G9fYvZqytp3gMnC4-no9EB=Jnzqmu44i8JQo6apiZat-xxPg@mail.gmail.com>
 <CAG_fn=V3a-kLkjE252V4ncHWDR0YhMby7nd1P6RNQA4aPf+fRw@mail.gmail.com> <CAG_fn=VuD+8GL_3-aSa9Y=zLqmroK11bqk48GBuPgTCpZMe-jw@mail.gmail.com>
In-Reply-To: <CAG_fn=VuD+8GL_3-aSa9Y=zLqmroK11bqk48GBuPgTCpZMe-jw@mail.gmail.com>
From:   Marco Elver <elver@google.com>
Date:   Fri, 17 Feb 2023 00:52:03 +0100
Message-ID: <CANpmjNOciiDNkWDrkQ+BEgAj=rSYGQAuHVS1DTDfvPHSbAndoA@mail.gmail.com>
Subject: Re: next: x86_64: kunit test crashed and kernel panic
To:     Alexander Potapenko <glider@google.com>
Cc:     Naresh Kamboju <naresh.kamboju@linaro.org>,
        Peter Zijlstra <peterz@infradead.org>,
        Jakub Jelinek <jakub@redhat.com>,
        Peter Collingbourne <pcc@google.com>,
        kasan-dev <kasan-dev@googlegroups.com>,
        open list <linux-kernel@vger.kernel.org>,
        kunit-dev@googlegroups.com, lkft-triage@lists.linaro.org,
        regressions@lists.linux.dev,
        Anders Roxell <anders.roxell@linaro.org>,
        Arnd Bergmann <arnd@arndb.de>,
        Andrey Konovalov <andreyknvl@gmail.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Thu, 16 Feb 2023 at 19:59, Alexander Potapenko <glider@google.com> wrote:
>
> >
> > > <4>[   38.796558]  ? kmalloc_memmove_negative_size+0xeb/0x1f0
> > > <4>[   38.797376]  ? __pfx_kmalloc_memmove_negative_size+0x10/0x10
> >
> > Most certainly kmalloc_memmove_negative_size() is related.
> > Looks like we fail to intercept the call to memmove() in this test,
> > passing -2 to the actual __memmove().
>
> This was introduced by 69d4c0d321869 ("entry, kasan, x86: Disallow
> overriding mem*() functions")

Ah, thanks!

> There's Marco's "kasan: Emit different calls for instrumentable
> memintrinsics", but it doesn't fix the problem for me (looking
> closer...), and GCC support is still not there, right?

Only Clang 15 supports it at this point. Some future GCC will support it.

> Failing to intercept memcpy/memset/memmove should normally result in
> false negatives, but kmalloc_memmove_negative_size() makes a strong
> assumption that KASAN will catch and prevent memmove(dst, src, -2).

Ouch - ok, so we need to skip these tests if we know memintrinsics
aren't instrumented.

I've sent a series here:
https://lore.kernel.org/all/20230216234522.3757369-1-elver@google.com/