From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:38982) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1e2B1S-0002fI-GG for qemu-devel@nongnu.org; Wed, 11 Oct 2017 03:01:43 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1e2B1P-0003OJ-BC for qemu-devel@nongnu.org; Wed, 11 Oct 2017 03:01:42 -0400 Received: from mx1.redhat.com ([209.132.183.28]:43092) by eggs.gnu.org with esmtps (TLS1.0:DHE_RSA_AES_256_CBC_SHA1:32) (Exim 4.71) (envelope-from ) id 1e2B1P-0003Mj-4W for qemu-devel@nongnu.org; Wed, 11 Oct 2017 03:01:39 -0400 Message-ID: <1507705293.31518.7.camel@redhat.com> From: Gerd Hoffmann Date: Wed, 11 Oct 2017 09:01:33 +0200 In-Reply-To: <20171010142706.360019aa@t450s.home> References: <20171010140334.8231-1-kraxel@redhat.com> <20171010140334.8231-7-kraxel@redhat.com> <20171010142706.360019aa@t450s.home> Content-Type: text/plain; charset="UTF-8" Mime-Version: 1.0 Content-Transfer-Encoding: quoted-printable Subject: Re: [Qemu-devel] [RfC PATCH 6/6] vfio/display: add dmabuf support (v15) List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Alex Williamson Cc: qemu-devel@nongnu.org, Tina Zhang , intel-gvt-dev@lists.freedesktop.org Hi, > Looks pretty straight forward otherwise.=C2=A0=C2=A0I like the LRU dmab= uf > freeing, but is it mostly for validating the interface?=C2=A0=C2=A0My > impression > is that we'd reach a steady state with a single plane and single > cursor so I wonder if keeping that cache really provides a noticeable > benefit. Really depends on what the guest is doing. I didn't do too much testing with various cache sizes and guests yet. > =C2=A0=C2=A0Perhaps for a Linux guest switching between vt and graphics > mode? fbcon/vt is one dmabuf. xorg I have seen using multiple primary plane dma-bufs, dunno why, doesn't change often. wayland does page-flipping on each rendered frame, which translates to very frequent dma-buf changes and I expect the caching will pay off here. cursor dma-buf with xorg changes each time the cursor shape changes. > =C2=A0=C2=A0It also seems like the primary and cursor will be fighting = for > the head of the list, so if we keep lists, maybe they should be per > plane type. Makes sense indeed. > This uses the x-display option for now, how do we move past that to > get > automatic configuration? I think a OnOffAuto display property would be the best choice. cheers, Gerd