From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Michael S. Tsirkin" Subject: Re: [PATCH 16/22] virtio_pci: use separate notification offsets for each vq. Date: Wed, 27 Mar 2013 13:25:35 +0200 Message-ID: <20130327112535.GE24243@redhat.com> References: <1363854584-25795-1-git-send-email-rusty@rustcorp.com.au> <1363854584-25795-17-git-send-email-rusty@rustcorp.com.au> <20130321101300.GA30493@redhat.com> <87wqt0du2e.fsf@rustcorp.com.au> <20130324201910.GA31631@redhat.com> <8738vjer43.fsf@rustcorp.com.au> <20130326193911.GA19251@redhat.com> <87ip4d4sef.fsf@rustcorp.com.au> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: Content-Disposition: inline In-Reply-To: <87ip4d4sef.fsf@rustcorp.com.au> List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: virtualization-bounces@lists.linux-foundation.org Errors-To: virtualization-bounces@lists.linux-foundation.org To: Rusty Russell Cc: hpa@zytor.com, virtualization@lists.linux-foundation.org List-Id: virtualization@lists.linuxfoundation.org On Wed, Mar 27, 2013 at 10:37:20AM +1030, Rusty Russell wrote: > "Michael S. Tsirkin" writes: > > On Mon, Mar 25, 2013 at 08:30:28PM +1030, Rusty Russell wrote: > >> Let's go back a level. Do we still need I/O bars at all now? Or can we > >> say "if you want hundreds of vqs, use mem bars"? > >> > >> hpa wanted the option to have either, but do we still want that? > > > > hpa says having both is required for BIOS, not just for speed with KVM. > > OK so the offset must not be applied to the I/O bar as you suggested. Aha. Yes, good idea. As for how large the offsets are, I am guessing we should either just say offset is vqn * X and data is vqn, or give hypervisors full flexibility with 32 bit offset and arbitrary data. 16 bit offsets seem neither here nor there ... Not a strong preference. > Since AFAICT I/O bars are deprecated, should we insist that there be a > memory bar, and the I/O bar is optional? Or just leave it entirely > undefined, and say there can be either or both? I would make the memory bar required and the I/O bar optional. Again not a strong preference. > I dislike the idea of BIOS code which assumed an I/O bar and thus won't > work with a compliant device which doesn't provide one. I'd prefer all > compliant drivers to work with all compliant devices. > > Cheers, > Rusty. In any case, the only thing we would want in the IO BAR is the notification. So we should add a way to control device configuration through PCI configuration. An offset/data pair will do the trick. -- MST