Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Fri, 26 Apr 2019 14:08:56 -0700
From:   Jakub Kicinski <jakub.kicinski@netronome.com>
To:     Andrew Lunn <andrew@lunn.ch>
Cc:     Esben Haabendal <esben@geanix.com>, netdev@vger.kernel.org,
        "David S. Miller" <davem@davemloft.net>,
        Michal Simek <michal.simek@xilinx.com>,
        Yang Wei <yang.wei9@zte.com.cn>,
        YueHaibing <yuehaibing@huawei.com>,
        Luis Chamberlain <mcgrof@kernel.org>,
        linux-arm-kernel@lists.infradead.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH 03/12] net: ll_temac: Fix support for 64-bit platforms
Message-ID: <20190426140856.78b786ee@cakuba.netronome.com>
In-Reply-To: <20190426205912.GI4041@lunn.ch>
References: <20190426073231.4008-1-esben@geanix.com>
        <20190426073231.4008-4-esben@geanix.com>
        <20190426114013.333e62c8@cakuba.netronome.com>
        <20190426205912.GI4041@lunn.ch>
Organization: Netronome Systems, Ltd.
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Fri, 26 Apr 2019 22:59:12 +0200, Andrew Lunn wrote:
> On Fri, Apr 26, 2019 at 11:40:13AM -0700, Jakub Kicinski wrote:
> > On Fri, 26 Apr 2019 09:32:22 +0200, Esben Haabendal wrote:  
> > > The use of buffer descriptor APP4 field (32-bit) for storing skb pointer
> > > obviously does not work on 64-bit platforms.
> > > As APP3 is also unused, we can use that to store the other half of 64-bit
> > > pointer values.
> > > 
> > > Contrary to what is hinted at in commit message of commit 15bfe05c8d63
> > > ("net: ethernet: xilinx: Mark XILINX_LL_TEMAC broken on 64-bit")
> > > there are no other pointers stored in cdmac_bd.
> > > 
> > > Signed-off-by: Esben Haabendal <esben@geanix.com>  
> > 
> > This is a bit strange, the driver stores the host's virtual address into
> > the HW descriptor?  
> 
> Hi Jukub

I need to start keeping track of all the ways my name gets spelled :)
I find it entertaining :)

> This is reasonably common. You need some sort of cookie which links
> the hardware descriptor to the skbuf it points to. The hardware makes
> no use of it, it is just a cookie.

Right, but accesses to HW descriptor memory ring are significantly 
more expensive, especially on platforms which are not coherent with 
DMA operations (everything but x86?)

A preferable design is to have two descriptor rings - one for HW
descriptors and one for software context, no?