netdev - Re: [RFC net-2.6.24] skbuff: change skb_frag

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] [day] [month] [year] [list]

Message-Id: <20071008.000451.38691636.davem@davemloft.net>
Date:	Mon, 08 Oct 2007 00:04:51 -0700 (PDT)
From:	David Miller <davem@...emloft.net>
To:	shemminger@...ux-foundation.org
Cc:	netdev@...r.kernel.org
Subject: Re: [RFC net-2.6.24] skbuff: change skb_frag_struct to scatterlist

From: Stephen Hemminger <shemminger@...ux-foundation.org>
Date: Fri, 5 Oct 2007 17:43:22 -0700

> Replace the skb frag list with the common scatterlist definition.
> This allows device drivers to use dma_scatter/gather operations which
> may be faster on some platforms. As a side benefit, it is easier to
> handle dma mapping error unwind.
> 
> This idea came up long ago, just never got implemented.
> Reimplemented against net-2.6.24.  This version is for comment, not
> tested yet.
> 
> Signed-off-by: Stephen Hemminger <shemminger@...ux-foundation.org>

Thanks for keeping this idea alive.

But this isn't the hard part.  The issue is that skb->data is still
seperate, and as long as that's true DMA mapping error recovery is
still a complete mess because we still have to do two seperate DMA
mappings on transmit, one of which is conditional, instead of one call
which is easy to handle properly.

We could do something cute like have the packet building routines use
skb_shinfo(skb)->frags[1] and onward, and then right before
->hard_start_xmit() we fill in skb_shinfo(skb)->frags[0] with
virt_to_page(skb->data) etc.

No page refcount bumps on this magic first entry, in order for it to
be as near zero cost as possible.

Anyways, just one idea.
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html