Date: Wed, 2 Jun 2004 03:30:27 -0700
From: Mitchell Blank Jr <mitch@sfgoth.com>
To: "Ihar 'Philips' Filipau" <ifilipau@giga-stream.de>
Cc: netdev@oss.sgi.com,
       Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Subject: Re: e1000 question
Message-ID: <20040602103027.GA74881@gaz.sfgoth.com>
References: <40BD8E49.3070605@giga-stream.de>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <40BD8E49.3070605@giga-stream.de>
User-Agent: Mutt/1.4.2.1i
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1059
Lines: 23

Ihar 'Philips' Filipau wrote:
>    Functions e1000_clean_{t,r}x_irq are very similar: both of them are 
> checking descriptor flag updated by nic.
>    Host CPU, obviously, to perform this check, will cache descriptor.
>    If, say e1000_clean_rx_irq() will be called twice in short time 
> range, I expect that it can miss change of the flag, since old flag may 
> still sit in host CPU cache.

Please see Documentation/DMA-mapping.txt; especially the part starting
at "There are two types of DMA mappings..."  Ring buffers are allocated
as "consistent" DMA memory.

For most architectures this will mean that the cache hardware snoops the
PCI bus and automatically invalidates cache lines as they are written to.
For architectures that can't do that then Linux will mark those memory
regions uncacheable.

-Mitch
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/