From mail at lechevalier.se  Thu Jul 23 17:36:51 2020
From: mail at lechevalier.se (A L)
Date: Thu, 23 Jul 2020 17:36:51 +0200
Subject: [gnutls-help] Support AF_ALG crypto in Linux
Message-ID: <21d489d9-bc32-39c5-fdc3-e8f71c100817@lechevalier.se>

Hi,

These days all AMD CPUs have a on-board Crypotographic CoProcessor 
(CCP), which is part of the Platform Security Processor (PSP). It has 
hardware encryption and decryption as well as a randnr generator. In 
current Kernels, the CCP is available throuh AF_ALG interface.

I do not think that GnuTLS supports this. Doing a web search I found 
only one reference to a 2 year old patch: 
https://gitlab.com/gnutls/gnutls/-/issues/308

IMHO it would be great if GnuTLS would consider adding support for this. 
On my AMD Athlon 3000G, OpenSSL generated almost 30x performance 
increase with AF_ALG enabled!

Relevant kernel doc: 
https://www.kernel.org/doc/html/v5.7/crypto/userspace-if.html

Anders


From ueno at gnu.org  Mon Jul 27 09:38:17 2020
From: ueno at gnu.org (Daiki Ueno)
Date: Mon, 27 Jul 2020 09:38:17 +0200
Subject: [gnutls-help] Support AF_ALG crypto in Linux
In-Reply-To: <21d489d9-bc32-39c5-fdc3-e8f71c100817@lechevalier.se> (A. L.'s
 message of "Thu, 23 Jul 2020 17:36:51 +0200")
References: <21d489d9-bc32-39c5-fdc3-e8f71c100817@lechevalier.se>
Message-ID: <87sgdd9zme.fsf-ueno@gnu.org>

Hello,

A L <mail at lechevalier.se> writes:

> I do not think that GnuTLS supports this. Doing a web search I found
> only one reference to a 2 year old patch: 
> https://gitlab.com/gnutls/gnutls/-/issues/308

Let's add Stephan in the loop.  I would be happy to review once the
patches are submitted as a normal merge request.

> IMHO it would be great if GnuTLS would consider adding support for
> this. On my AMD Athlon 3000G, OpenSSL generated almost 30x performance 
> increase with AF_ALG enabled!

Just for curiousity, what algorithm did you test?

Regards,
-- 
Daiki Ueno


From mail at lechevalier.se  Tue Jul 28 12:38:09 2020
From: mail at lechevalier.se (A L)
Date: Tue, 28 Jul 2020 12:38:09 +0200
Subject: [gnutls-help] Support AF_ALG crypto in Linux
In-Reply-To: <87sgdd9zme.fsf-ueno@gnu.org>
References: <21d489d9-bc32-39c5-fdc3-e8f71c100817@lechevalier.se>
 <87sgdd9zme.fsf-ueno@gnu.org>
Message-ID: <c7309ccd-7cde-840b-3a82-4f106cfccaa2@lechevalier.se>



On 2020-07-27 09:38, Daiki Ueno wrote:
> Hello,
>
> A L <mail at lechevalier.se> writes:
>
>> I do not think that GnuTLS supports this. Doing a web search I found
>> only one reference to a 2 year old patch:
>> https://gitlab.com/gnutls/gnutls/-/issues/308
> Let's add Stephan in the loop.  I would be happy to review once the
> patches are submitted as a normal merge request.
>
>> IMHO it would be great if GnuTLS would consider adding support for
>> this. On my AMD Athlon 3000G, OpenSSL generated almost 30x performance
>> increase with AF_ALG enabled!
> Just for curiousity, what algorithm did you test?
>
> Regards,

I used the aes-192-cbc for benchmark. Others like sha256 work too, but 
does not yield as big of an improvement on this particular CPU.


# openssl speed -evp aes-192-cbc -engine afalg
Code:
engine "afalg" set.
Doing aes-192-cbc for 3s on 16 size blocks: 1685326 aes-192-cbc's in 0.45s
Doing aes-192-cbc for 3s on 64 size blocks: 1722473 aes-192-cbc's in 0.41s
Doing aes-192-cbc for 3s on 256 size blocks: 1543359 aes-192-cbc's in 0.40s
Doing aes-192-cbc for 3s on 1024 size blocks: 1127194 aes-192-cbc's in 0.33s
Doing aes-192-cbc for 3s on 8192 size blocks: 335502 aes-192-cbc's in 0.09s
Doing aes-192-cbc for 3s on 16384 size blocks: 180981 aes-192-cbc's in 0.06s
OpenSSL 1.1.1g? 21 Apr 2020
built on: Thu Jul 23 11:19:52 2020 UTC
options:bn(64,64) rc4(8x,int) des(int) aes(partial) idea(int) blowfish(ptr)
compiler: x86_64-pc-linux-gnu-gcc -fPIC -pthread -m64 -Wa,--noexecstack 
-O2 -march=native -pipe -fno-strict-aliasing -Wa,--noexecstack 
-DOPENSSL_USE_NODELETE -DL_ENDIAN -DOPENSSL_PIC -DOPENSSL_CPUID_OBJ 
-DOPENSSL_IA32_SSE2 -DOPENSSL_BN_ASM_MONT -DOPENSSL_BN_ASM_MONT5 
-DOPENSSL_BN_ASM_GF2m -DSHA1_ASM -DSHA256_ASM -DSHA512_ASM 
-DKECCAK1600_ASM -DRC4_ASM -DMD5_ASM -DAESNI_ASM -DVPAES_ASM -DGHASH_ASM 
-DECP_NISTZ256_ASM -DX25519_ASM -DPOLY1305_ASM -DZLIB -DNDEBUG 
-DOPENSSL_NO_BUF_FREELISTS
The 'numbers' are in 1000s of bytes per second processed.
type???????????? 16 bytes???? 64 bytes??? 256 bytes?? 1024 bytes?? 8192 
bytes? 16384 bytes
aes-192-cbc????? 59922.70k?? 268873.83k?? 987749.76k 3497717.14k 
30538137.60k 49419878.40k


# openssl speed -evp aes-192-cbc
Code:
Doing aes-192-cbc for 3s on 16 size blocks: 139159126 aes-192-cbc's in 2.99s
Doing aes-192-cbc for 3s on 64 size blocks: 51864313 aes-192-cbc's in 2.99s
Doing aes-192-cbc for 3s on 256 size blocks: 13886330 aes-192-cbc's in 2.99s
Doing aes-192-cbc for 3s on 1024 size blocks: 3540324 aes-192-cbc's in 3.00s
Doing aes-192-cbc for 3s on 8192 size blocks: 444244 aes-192-cbc's in 2.99s
Doing aes-192-cbc for 3s on 16384 size blocks: 222334 aes-192-cbc's in 2.99s
OpenSSL 1.1.1g? 21 Apr 2020
built on: Thu Jul 23 11:19:52 2020 UTC
options:bn(64,64) rc4(8x,int) des(int) aes(partial) idea(int) blowfish(ptr)
compiler: x86_64-pc-linux-gnu-gcc -fPIC -pthread -m64 -Wa,--noexecstack 
-O2 -march=native -pipe -fno-strict-aliasing -Wa,--noexecstack 
-DOPENSSL_USE_NODELETE -DL_ENDIAN -DOPENSSL_PIC -DOPENSSL_CPUID_OBJ 
-DOPENSSL_IA32_SSE2 -DOPENSSL_BN_ASM_MONT -DOPENSSL_BN_ASM_MONT5 
-DOPENSSL_BN_ASM_GF2m -DSHA1_ASM -DSHA256_ASM -DSHA512_ASM 
-DKECCAK1600_ASM -DRC4_ASM -DMD5_ASM -DAESNI_ASM -DVPAES_ASM -DGHASH_ASM 
-DECP_NISTZ256_ASM -DX25519_ASM -DPOLY1305_ASM -DZLIB -DNDEBUG 
-DOPENSSL_NO_BUF_FREELISTS
The 'numbers' are in 1000s of bytes per second processed.
type???????????? 16 bytes???? 64 bytes??? 256 bytes?? 1024 bytes?? 8192 
bytes? 16384 bytes
aes-192-cbc???? 744664.22k? 1110139.14k? 1188929.93k 1208430.59k? 
1217139.41k? 1218301.09k