Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Tue, 19 Jul 2022 16:11:50 +0000
From:   Keir Fraser <keirf@google.com>
To:     Christoph Hellwig <hch@infradead.org>
Cc:     "Michael S. Tsirkin" <mst@redhat.com>,
        Jason Wang <jasowang@redhat.com>, kernel-team@android.com,
        virtualization@lists.linux-foundation.org,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH] virtio: Force DMA restricted devices through DMA API
Message-ID: <YtbXxl8STUtQkacb@google.com>
References: <20220719100256.419780-1-keirf@google.com>
 <YtbMcBw4l0LAFn9+@infradead.org>
 <YtbRwO40CmIRWOUR@google.com>
 <YtbTGjRw65QEKkQA@infradead.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <YtbTGjRw65QEKkQA@infradead.org>
Precedence: bulk

On Tue, Jul 19, 2022 at 08:51:54AM -0700, Christoph Hellwig wrote:
> On Tue, Jul 19, 2022 at 03:46:08PM +0000, Keir Fraser wrote:
> > However, if the general idea at least is acceptable, would the
> > implementation be acceptable if I add an explicit API for this to the
> > DMA subsystem, and hide the detail there?
> 
> I don't think so.  The right thing to key off is
> VIRTIO_F_ACCESS_PLATFORM, which really should be set in any modern
> virtio device after all the problems we had with the lack of it.

Ok. Certainly the flag description in virtio spec fits the bill.

> > Or a completely different approach would be to revert the patch
> > e41b1355508d which clears VIRTIO_F_ACCESS_PLATFORM in the balloon
> > driver. MST: That's back in your court, as it's your patch!
> 
> Which also means this needs to be addresses, but I don't think a
> simple revert is enough.

Well here are two possible approaches:

1. Revert e41b1355508d outright. I'm not even sure what it would mean
for reported pages to go through IOMMU. And VIRTIO_F_ACCESS_PLATFORM
is no longer IOMMU-specific anyway.

2. Continue to clear the flag during virtio_balloon negotiation, but
remember that it was offered, and test for that in vring_use_dma_api()
as well as, or instead of, virtio_has_dma_quirk().

Do either of those appeal?