From: will.deacon@arm.com (Will Deacon) Date: Mon, 13 Feb 2017 11:57:59 +0000 Subject: [PATCH] virtio: Try to untangle DMA coherency In-Reply-To: <20170210191409-mutt-send-email-mst@kernel.org> References: <20170201195732-mutt-send-email-mst@kernel.org> <20170201182659.GM8177@arm.com> <20170201210648-mutt-send-email-mst@kernel.org> <20170202112614.GB30577@arm.com> <20170202182224-mutt-send-email-mst@kernel.org> <20170202164049.GI13839@arm.com> <20170209201341-mutt-send-email-mst@kernel.org> <20170209183118.GI19397@arm.com> <20170210191409-mutt-send-email-mst@kernel.org> Message-ID: <20170213115758.GK1512@arm.com> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org On Fri, Feb 10, 2017 at 07:16:10PM +0200, Michael S. Tsirkin wrote: > On Thu, Feb 09, 2017 at 06:31:18PM +0000, Will Deacon wrote: > > On ARM (and other archs such as > > Power), having a mismatch between a cacheable and a non-cacheable mapping > > can result in a loss of coherency between the two (for example, if the > > non-cacheable gues accesses bypass the cache, but the cacheable host > > accesses allocate in the cache). > > I guess it's an optimization to avoid cache snoops for non-cacheable accesses? The architecture doesn't rationalise the decision, but a micro-architecture could indeed implement the optimisation you suggest (and we do observe the loss of coherency in practice). Will