netdev - Re: [PATCH] net: mvpp2: fix dma unmapping of TX buffers for fragments

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: Windows password security audit tool. GUI, reports in PDF.

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

Date:   Sat, 17 Dec 2016 10:20:57 -0500 (EST)
From:   David Miller <davem@...emloft.net>
To:     thomas.petazzoni@...e-electrons.com
Cc:     netdev@...r.kernel.org, linux-arm-kernel@...ts.infradead.org,
        jason@...edaemon.net, andrew@...n.ch,
        sebastian.hesselbarth@...il.com,
        gregory.clement@...e-electrons.com, mw@...ihalf.com,
        stefanc@...vell.com, nadavh@...vell.com, hannah@...vell.com,
        yehuday@...vell.com, raphael.glon@...p.ovh.com,
        stable@...r.kernel.org
Subject: Re: [PATCH] net: mvpp2: fix dma unmapping of TX buffers for
 fragments

From: Thomas Petazzoni <thomas.petazzoni@...e-electrons.com>
Date: Tue, 13 Dec 2016 17:53:15 +0100

> diff --git a/drivers/net/ethernet/marvell/mvpp2.c b/drivers/net/ethernet/marvell/mvpp2.c
> index 1026c45..d168b13 100644
> --- a/drivers/net/ethernet/marvell/mvpp2.c
> +++ b/drivers/net/ethernet/marvell/mvpp2.c
> @@ -791,6 +791,8 @@ struct mvpp2_txq_pcpu {
>  	/* Array of transmitted buffers' physical addresses */
>  	dma_addr_t *tx_buffs;
>  
> +	size_t *tx_data_size;
> +

You're really destroying cache locality, and making things overly
complicated, by having two arrays.  Actually this is now the third in
this structure alone.  That's crazy.

Just have one array for the TX ring software state:

struct tx_buff_info {
	struct sk_buff	*skb;
	dma_addr_t	dma_addr;
	unsigned int	size;
};

Then in the per-cpu TX struct:

	struct tx_buff_info	*info;

This way every data access by the cpu for processing a ring entry
will be localized, increasing cache hit rates.

This also significantly simplifies the code that allocates and
frees this memory.