From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Hans J. Koch" Subject: Re: [PATCH] uio_pci_generic does not export memory resources Date: Fri, 8 Jun 2012 18:37:29 +0200 Message-ID: <20120608163728.GD9705@local> References: <1339156616.3870.9.camel@blech> <20120608130351.GB1964@redhat.com> <1339165738.3870.18.camel@blech> <20120608151843.GA9705@local> <1339170312.26423.6.camel@tesla> <20120608155728.GB9705@local> <1339172599.26423.13.camel@tesla> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: "Hans J. Koch" , "Michael S. Tsirkin" , Greg Kroah-Hartman , kvm@vger.kernel.org, linux-kernel@vger.kernel.org To: Dominic Eschweiler Return-path: Content-Disposition: inline In-Reply-To: <1339172599.26423.13.camel@tesla> Sender: linux-kernel-owner@vger.kernel.org List-Id: kvm.vger.kernel.org On Fri, Jun 08, 2012 at 06:23:19PM +0200, Dominic Eschweiler wrote: > Am Freitag, den 08.06.2012, 17:57 +0200 schrieb Hans J. Koch: > > What problem do you have with this approach? > > Nothing itself. There is obviously a documentation issue, which really > confuses people who are new to that stuff. Well, the documentation says "Userspace driver can use pci sysfs interface, or the libpci libray that wraps it, to talk to the device..." I agree that this is not very clear, especially since the example code there doesn't use this. > I thought that fixing it, by > just adding a view lines of code (really a view, when I remove the DMA > masque part) might enhance this particular module. I'm not sure if talking to the device the UIO way can confuse the PCI subsystem. Are you? Any hints from somebody else? Unfortunately, I can't remember why we agreed to do it the way it is... > > Anyway, what is wrong when the uio_pci_generic module works like > described in the related documentation? It does, the documentation is just not very explicit when it comes to memory mapping. Thanks, Hans