Received-SPF: pass (google.com: domain of linux-crypto-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
From:   Linus Torvalds <torvalds@linux-foundation.org>
Date:   Wed, 28 Jun 2023 20:13:25 -0700
Message-ID: <CAHk-=whXn0YTojV=+J8B-r8KLvNtqc2JtCa4a_bdhf+=GN5OOw@mail.gmail.com>
Subject: Build error in crypto/marvell/cesa/cipher.c
To:     Kees Cook <keescook@chromium.org>,
        Boris Brezillon <bbrezillon@kernel.org>,
        Arnaud Ebalard <arno@natisbad.org>,
        Srujana Challa <schalla@marvell.com>,
        Mustafa Ismail <mustafa.ismail@intel.com>,
        Shiraz Saleem <shiraz.saleem@intel.com>
Cc:     Herbert Xu <herbert@gondor.apana.org.au>,
        Linux Crypto Mailing List <linux-crypto@vger.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Jason Gunthorpe <jgg@ziepe.ca>,
        Leon Romanovsky <leon@kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

So I don't see anything that has changed, and I suspect the only
change is that my compiler version changed, but my arm64 build fails
right now with FORTIFY_STRING enabled.

On arm64 I now get this warning:

  In function 'fortify_memcpy_chk',
    inlined from 'mv_cesa_des3_ede_setkey' and
drivers/crypto/marvell/cesa/cipher.c:307:2:
  ./include/linux/fortify-string.h:583:25: error: call to
'__write_overflow_field' declared with attribute warning: detected
write beyond size of field (1st parameter); maybe use struct_group()?
[-Werror=attribute-warning[

but I haven't been compiling regularly enough to know when this
warning suddenly started showing up.

I enabled the cesa driver on x86-64 (by also letting it build with
COMPILE_TEST), and I do *not* see this warning on x86-64, which makes
me think it's the compiler version that matters here.

On my arm64 setup, I have gcc-13.1.1, while my x86-64 build is still 12.3.1.

But I think the warning is correct.  The problem is that the 'ctx'
pointer is wrongly typed, and it's using "struct mv_cesa_des_ctx"
(which has a "key[]" size of DES_KEY_SIZE).

And it *should* use "struct mv_cesa_des3_ctx" which has otherwise the
same layout, but the right key size (DES3_EDE_KEY_SIZE).

Fixing that type fixes the warning.

I'm actually surprised that gcc-12 doesn't seem to warn about this.
Kees? This looks like a rather obvious overflow, which makes me think
I'm missing something.

I get a similar error in 'irdma_clr_wqes()' at
drivers/infiniband/hw/irdma/uk.c:103 (and same thing on line 105). I
don't see what the right solution there is, but it looks like we have

        IRDMA_CQP_WQE_SIZE = 8
        __le64 elem[IRDMA_CQP_WQE_SIZE];

and it's doing a 4kB memset to that element. The mistake is not as
obvious as in the cesa driver.

Kees, any idea why I'm seeing it now? Is it the new
-fstrict-flex-arrays=3? And if so, why? None of this is about flex
arrays...

Anyway, please forward me fixes so that I can have a working arm64
test build again....

               Linus