[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAJKOXPdveQeg_M42v69m9vBEz2SaJE0_BZNRUqGLt4Ve+ZLYSg@mail.gmail.com>
Date: Tue, 24 Jul 2018 13:11:48 +0200
From: Krzysztof Kozlowski <krzk@...nel.org>
To: David Laight <David.Laight@...lab.com>
Cc: Pantelis Antoniou <pantelis.antoniou@...il.com>,
"David S. Miller" <davem@...emloft.net>,
"linuxppc-dev@...ts.ozlabs.org" <linuxppc-dev@...ts.ozlabs.org>,
"netdev@...r.kernel.org" <netdev@...r.kernel.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
Eric Biggers <ebiggers3@...il.com>
Subject: Re: [PATCH] net: ethernet: fs-enet: Use generic CRC32 implementation
On 24 July 2018 at 13:05, David Laight <David.Laight@...lab.com> wrote:
> From: Krzysztof Kozlowski
>> Sent: 23 July 2018 17:20
>> Use generic kernel CRC32 implementation because it:
>> 1. Should be faster (uses lookup tables),
>
> Are you sure?
> The lookup tables are unlikely to be in the data cache and
> the 6 cache misses kill performance.
> (Not that it particularly matters when setting up multicast hash tables).
Good point, so this statement should be rather "Could be faster"... I
did not run any performance tests so this is not backed up by any
data.
I think the main benefit is rather easier code maintenance by removing
duplicated, custom code.
>> 2. Removes duplicated CRC generation code,
>> 3. Uses well-proven algorithm instead of coding it one more time.
> ...
>>
>> Not tested on hardware.
>
> Have you verified that the old and new functions give the
> same result for a few mac addresses?
> It is very easy to use the wrong bits in crc calculations
> or generate the output in the wrong bit order.
I copied the original code and new one onto a different driver and run
this in a loop for thousands of data input (although not all possible
MAC combinations). The output was the same. I agree however that real
testing would be important.
Best regards,
Krzysztof
Powered by blists - more mailing lists