linux-kernel - Re: [PATCH 08/12] crypto: caam - support crypto

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <VI1PR04MB444537DA6774BE5E31F2C04F8C490@VI1PR04MB4445.eurprd04.prod.outlook.com>
Date:   Fri, 22 Nov 2019 14:11:46 +0000
From:   Iuliana Prodan <iuliana.prodan@....com>
To:     Herbert Xu <herbert@...dor.apana.org.au>
CC:     Horia Geanta <horia.geanta@....com>,
        Aymen Sghaier <aymen.sghaier@....com>,
        "David S. Miller" <davem@...emloft.net>,
        Tom Lendacky <thomas.lendacky@....com>,
        Gary Hook <gary.hook@....com>,
        "linux-crypto@...r.kernel.org" <linux-crypto@...r.kernel.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        dl-linux-imx <linux-imx@....com>
Subject: Re: [PATCH 08/12] crypto: caam - support crypto_engine framework for
 SKCIPHER algorithms

On 11/22/2019 1:09 PM, Herbert Xu wrote:
> On Fri, Nov 22, 2019 at 11:05:59AM +0000, Iuliana Prodan wrote:
>>
>> Sorry, but I don't understand what is wrong here? I'm using the correct
>> type, the specific type, for the request when sending it to crypto engine.
>> This transfer_request_to_engine function is called from caam_jr_enqueue,
>> where I have all types of request, so I'm using the async_request, and
>> when transferring to crypto engine I cast it to the specific type.
> 
> These internal types are only for use by the crypto API and helper
> code such as crypto_engine.  They should not be used by drivers in
> general.
> 
So, just to be clear, I shouldn't use crypto_async_request in driver code?
I see that this generic crypto request is used in multiple drivers.

>> I believe is an overhead to sent all request to crypto engine since most
>> of them can be directly executed by hw.
>> Also, in case there is no need for backlog and the hw is busy we can
>> drop the request.
> 
> If the crypto_engine has so much overhead then you should work on
> fixing crypto_engine and not work around it like this.
> 
I can try sending _all_ requests to crypto engine and make some 
performance measurements to see which solution is best.

Thanks,
Iulia