Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 25 Nov 2016 17:00:10 +0100
From:      Mateusz Guzik <mjguzik@gmail.com>
To:        Fabien Thomas <fabient@FreeBSD.org>
Cc:        src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org
Subject:   Re: svn commit: r309143 - head/sys/libkern
Message-ID:  <20161125160010.GA3307@dft-labs.eu>
In-Reply-To: <201611251349.uAPDnX09042110@repo.freebsd.org>
References:  <201611251349.uAPDnX09042110@repo.freebsd.org>

next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, Nov 25, 2016 at 01:49:33PM +0000, Fabien Thomas wrote:
> Author: fabient
> Date: Fri Nov 25 13:49:33 2016
> New Revision: 309143
> URL: https://svnweb.freebsd.org/changeset/base/309143
> 
> Log:
>   In a dual processor system (2*6 cores) during IPSec throughput tests,
>   we see a lot of contention on the arc4 lock, used to generate the IV
>   of the ESP output packets.
>   
>   The idea of this patch is to split this mutex in order to reduce the
>   contention on this lock.
>   
> +MALLOC_DEFINE(M_ARC4RANDOM, "arc4random", "arc4random structures");
>  
> -static u_int8_t arc4_randbyte(void);
> +struct arc4_s {
> +	u_int8_t i, j;
> +	int numruns;
> +	u_int8_t sbox[256];
> +	time_t t_reseed;
> +
> +	struct mtx mtx;
> +};
> +

Why is the lock at the end? Right now you got false sharing with the
next entry.

That said, I suggest:
1. moving thelock to the beginning
2. annotating the struct with __aligned(CACHE_LINE_SIZE)

-- 
Mateusz Guzik <mjguzik gmail.com>



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20161125160010.GA3307>