Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
To:     Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc:     "Martin K. Petersen" <martin.petersen@oracle.com>,
        Herbert Xu <herbert@gondor.apana.org.au>,
        Jeffrey Lien <Jeff.Lien@wdc.com>,
        David Darrington <david.darrington@wdc.com>,
        Christoph Hellwig <hch@infradead.org>,
        Jeff Furlong <jeff.furlong@wdc.com>,
        linux-block@vger.kernel.org,
        "open list\:HARDWARE RANDOM NUMBER GENERATOR CORE" 
        <linux-crypto@vger.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        linux-scsi@vger.kernel.org, Tim Chen <tim.c.chen@linux.intel.com>
Subject: Re: [PATCH 4/4] block: Integrity profile init function to trigger module loads
From:   "Martin K. Petersen" <martin.petersen@oracle.com>
Organization: Oracle Corporation
References: <20180825061205.ygrjjazkooqghrqy@gondor.apana.org.au>
        <20180826024006.13800-1-martin.petersen@oracle.com>
        <20180826024006.13800-4-martin.petersen@oracle.com>
        <CAKv+Gu92yj5Cuz36bRS7wz3Lg2srp8OFJC6QVO0UWYSjm8eEeg@mail.gmail.com>
Date:   Sun, 26 Aug 2018 09:30:39 -0400
In-Reply-To: <CAKv+Gu92yj5Cuz36bRS7wz3Lg2srp8OFJC6QVO0UWYSjm8eEeg@mail.gmail.com>
        (Ard Biesheuvel's message of "Sun, 26 Aug 2018 09:22:43 +0100")
Message-ID: <yq1h8jh6yrk.fsf@oracle.com>
User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/25.3 (gnu/linux)
MIME-Version: 1.0
Content-Type: text/plain
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


Hi Ard,

>> However, since the dependency for "crc10dif" is already satisfied,
>> nothing is going to cause the hardware-accelerated kernel modules to
>> get loaded.
>
> This is not true. All accelerated implementations based on SIMD
> polynomial multiplication are tried to the respective CPU features
> bits. This applies to x86, power, ARM and arm64.
>
> E.g., for x86 you have
>
> alias:          cpu:type:x86,ven*fam*mod*:feature:*0081*
>
> which will be matched by udev if /sys/devices/system/cpu/modalias
> contains feature 0081, and so the modules will be loaded automatically
> at boot.

If I can avoid carrying that init callback in the block integrity code
that will definitely make me happy. However, loading crct10dif-pclmul
does not happen automatically for me. crc-t10dif is linked statically
and every user of the CRC goes through that library. So nothing ever
requests the "crct10dif" modalias and no accelerator modules are loaded.

<fresh boot>

# lsmod | grep crc
crc32c_intel           24576  0
crc_ccitt              16384  1 ipv6

# modinfo crc32c_intel | grep cpu:type
alias:          cpu:type:x86,ven*fam*mod*:feature:*0094*

# modinfo crct10dif-pclmul | grep cpu:type
alias:          cpu:type:x86,ven*fam*mod*:feature:*0081*

# egrep -o "0081|0094" /sys/devices/system/cpu/modalias
0081
0094

# modprobe crct10dif
# lsmod | grep crc
crct10dif_pclmul       16384  1
crc32c_intel           24576  0
crc_ccitt              16384  1 ipv6

It's interesting that crc32c_intel is loaded but libcrc32c is not. That
matches your description of how things should work. But crct10dif-pclmul
isn't loaded and neither is crc32_pclmul:

# modprobe crc32
# lsmod | grep crc
crc32_generic          16384  0
crc32_pclmul           16384  0
crc32c_intel           24576  0
crc_ccitt              16384  1 ipv6

-- 
Martin K. Petersen	Oracle Linux Engineering