Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.33 as permitted sender) client-ip=23.128.96.33;
Message-ID: <b5e1427715aa20736791cc9378935dfd5e917ebc.camel@linux.ibm.com>
Subject: Re: [PATCH v2 1/2] iommu/virtio: Make use of ops->iotlb_sync_map
From:   Niklas Schnelle <schnelle@linux.ibm.com>
To:     Robin Murphy <robin.murphy@arm.com>,
        Jean-Philippe Brucker <jean-philippe@linaro.org>,
        Joerg Roedel <joro@8bytes.org>, Will Deacon <will@kernel.org>
Cc:     virtualization@lists.linux-foundation.org, iommu@lists.linux.dev,
        linux-kernel@vger.kernel.org
Date:   Tue, 19 Sep 2023 10:00:08 +0200
In-Reply-To: <ae7e513b-eb86-97e2-bed0-3cca91b8c959@arm.com>
References: <20230918-viommu-sync-map-v2-0-f33767f6cf7a@linux.ibm.com>
         <20230918-viommu-sync-map-v2-1-f33767f6cf7a@linux.ibm.com>
         <ae7e513b-eb86-97e2-bed0-3cca91b8c959@arm.com>
Content-Type: text/plain; charset="UTF-8"
User-Agent: Evolution 3.48.4 (3.48.4-1.fc38) 
Content-Transfer-Encoding: quoted-printable
MIME-Version: 1.0
Precedence: bulk

On Mon, 2023-09-18 at 17:37 +0100, Robin Murphy wrote:
> On 2023-09-18 12:51, Niklas Schnelle wrote:
> > Pull out the sync operation from viommu_map_pages() by implementing
> > ops->iotlb_sync_map. This allows the common IOMMU code to map multiple
> > elements of an sg with a single sync (see iommu_map_sg()). Furthermore,
> > it is also a requirement for IOMMU_CAP_DEFERRED_FLUSH.
>=20
> Is it really a requirement? Deferred flush only deals with unmapping. Or=
=20
> are you just trying to say that it's not too worthwhile to try doing=20
> more for unmapping performance while obvious mapping performance is=20
> still left on the table?
>=20

You're right there is no hard requirement. I somehow thought that
iommu_create_device_direct_map() relied on it because it does
flush_iotlb_all() and iommu_map() but that doesn't seem to be true. If
you want I can resend with the last sentence removed.

> > Link: https://lore.kernel.org/lkml/20230726111433.1105665-1-schnelle@li=
nux.ibm.com/
> > Signed-off-by: Niklas Schnelle <schnelle@linux.ibm.com>
> > ---
> >   drivers/iommu/virtio-iommu.c | 17 ++++++++++++++++-
> >   1 file changed, 16 insertions(+), 1 deletion(-)
> >=20
> > diff --git a/drivers/iommu/virtio-iommu.c b/drivers/iommu/virtio-iommu.c
> > index 17dcd826f5c2..3649586f0e5c 100644
> > --- a/drivers/iommu/virtio-iommu.c
> > +++ b/drivers/iommu/virtio-iommu.c
> > @@ -189,6 +189,12 @@ static int viommu_sync_req(struct viommu_dev *viom=
mu)
> >   	int ret;
> >   	unsigned long flags;
> >=20=20=20
> > +	/*
> > +	 * .iotlb_sync_map and .flush_iotlb_all may be called before the viom=
mu
> > +	 * is initialized e.g. via iommu_create_device_direct_mappings()
> > +	 */
> > +	if (!viommu)
> > +		return 0;
>=20
> Minor nit: I'd be inclined to make that check explicitly in the places=20
> where it definitely is expected, rather than allowing *any* sync to=20
> silently do nothing if called incorrectly. Plus then they could use=20
> vdomain->nr_endpoints for consistency with the equivalent checks=20
> elsewhere (it did take me a moment to figure out how we could get to=20
> .iotlb_sync_map with a NULL viommu without viommu_map_pages() blowing up=
=20
> first...)
>=20
> Thanks,
> Robin.

That's what I had in v1. I think this is a matter of taste and Jean-
Philippe pointed me to moving the check into viommu_sync_req() I added
a comment because it really is not entirely obvious.

>=20
> >   	spin_lock_irqsave(&viommu->request_lock, flags);
> >   	ret =3D __viommu_sync_req(viommu);
> >   	if (ret)
> > @@ -843,7 +849,7 @@ static int viommu_map_pages(struct iommu_domain *do=
main, unsigned long iova,
> >   			.flags		=3D cpu_to_le32(flags),
> >   		};
> >=20=20=20
> > -		ret =3D viommu_send_req_sync(vdomain->viommu, &map, sizeof(map));
> > +		ret =3D viommu_add_req(vdomain->viommu, &map, sizeof(map));
> >   		if (ret) {
> >   			viommu_del_mappings(vdomain, iova, end);
> >   			return ret;
> > @@ -912,6 +918,14 @@ static void viommu_iotlb_sync(struct iommu_domain =
*domain,
> >   	viommu_sync_req(vdomain->viommu);
> >   }
> >=20=20=20
> > +static int viommu_iotlb_sync_map(struct iommu_domain *domain,
> > +				 unsigned long iova, size_t size)
> > +{
> > +	struct viommu_domain *vdomain =3D to_viommu_domain(domain);
> > +
> > +	return viommu_sync_req(vdomain->viommu);
> > +}
> > +
> >   static void viommu_get_resv_regions(struct device *dev, struct list_h=
ead *head)
> >   {
> >   	struct iommu_resv_region *entry, *new_entry, *msi =3D NULL;
> > @@ -1058,6 +1072,7 @@ static struct iommu_ops viommu_ops =3D {
> >   		.unmap_pages		=3D viommu_unmap_pages,
> >   		.iova_to_phys		=3D viommu_iova_to_phys,
> >   		.iotlb_sync		=3D viommu_iotlb_sync,
> > +		.iotlb_sync_map		=3D viommu_iotlb_sync_map,
> >   		.free			=3D viommu_domain_free,
> >   	}
> >   };
> >=20