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;
Message-ID: <52fe7769ca5b66523c2c93c7d46ebc17dc144aca.camel@linux.ibm.com>
Subject: Re: [PATCH v2 4/7] iommu: Let iommu.strict override
 ops->def_domain_type
From:   Niklas Schnelle <schnelle@linux.ibm.com>
To:     Jason Gunthorpe <jgg@nvidia.com>
Cc:     Baolu Lu <baolu.lu@linux.intel.com>,
        Matthew Rosato <mjrosato@linux.ibm.com>,
        Gerd Bayer <gbayer@linux.ibm.com>, iommu@lists.linux.dev,
        Joerg Roedel <joro@8bytes.org>, Will Deacon <will@kernel.org>,
        Robin Murphy <robin.murphy@arm.com>,
        Wenjia Zhang <wenjia@linux.ibm.com>,
        Pierre Morel <pmorel@linux.ibm.com>,
        linux-s390@vger.kernel.org, borntraeger@linux.ibm.com,
        hca@linux.ibm.com, gor@linux.ibm.com,
        gerald.schaefer@linux.ibm.com, agordeev@linux.ibm.com,
        svens@linux.ibm.com, linux-kernel@vger.kernel.org,
        Julian Ruess <julianr@linux.ibm.com>
Date:   Mon, 28 Nov 2022 16:54:03 +0100
In-Reply-To: <Y4S3z6IpeDHmdUs/@nvidia.com>
References: <20221116171656.4128212-1-schnelle@linux.ibm.com>
         <20221116171656.4128212-5-schnelle@linux.ibm.com>
         <33eea9bd-e101-4836-19e8-d4b191b78b00@linux.intel.com>
         <9163440eb6a47fe02730638bbdf72fda5ee5ad2c.camel@linux.ibm.com>
         <Y4S3z6IpeDHmdUs/@nvidia.com>
Content-Type: text/plain; charset="UTF-8"
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit
Precedence: bulk

On Mon, 2022-11-28 at 09:29 -0400, Jason Gunthorpe wrote:
> On Mon, Nov 28, 2022 at 12:10:39PM +0100, Niklas Schnelle wrote:
> > On Thu, 2022-11-17 at 09:55 +0800, Baolu Lu wrote:
> > > On 2022/11/17 1:16, Niklas Schnelle wrote:
> > > > When iommu.strict=1 is set or iommu_set_dma_strict() was called we
> > > > should use IOMMU_DOMAIN_DMA irrespective of ops->def_domain_type.
> > > > 
> > > > Signed-off-by: Niklas Schnelle <schnelle@linux.ibm.com>
> > > > ---
> > > >   drivers/iommu/iommu.c | 3 +++
> > > >   1 file changed, 3 insertions(+)
> > > > 
> > > > diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
> > > > index 65a3b3d886dc..d9bf94d198df 100644
> > > > --- a/drivers/iommu/iommu.c
> > > > +++ b/drivers/iommu/iommu.c
> > > > @@ -1562,6 +1562,9 @@ static int iommu_get_def_domain_type(struct device *dev)
> > > >   {
> > > >   	const struct iommu_ops *ops = dev_iommu_ops(dev);
> > > >   
> > > > +	if (iommu_dma_strict)
> > > > +		return IOMMU_DOMAIN_DMA;
> > > 
> > > If any quirky device must work in IOMMU identity mapping mode, this
> > > might introduce functional regression. At least for VT-d platforms, some
> > > devices do require IOMMU identity mapping mode for functionality.
> > 
> > That's a good point. How about instead of unconditionally returning
> > IOMMU_DOMAIN_DMA we just do so if the domain type returned by ops-
> > > def_domain_type uses a flush queue (i.e. the __IOMMU_DOMAIN_DMA_FQ bit
> > is set). That way a device that only supports identity mapping gets to
> > set that but iommu_dma_strict at least always prevents use of an IOVA
> > flush queue.
> 
> I would prefer we create some formal caps in iommu_ops to describe
> whatever it is you are trying to do.
> 
> Jason

I agree that there is currently a lack of distinction between what
domain types can be used (capability) and which should be used as
default (iommu.strict=<x>, iommu_set_...(), CONFIG_IOMMU_DEFAULT_DMA,
ops->def_domain_type.).

My case though is about the latter which I think has some undocumented
and surprising precedences built in at the moment. With this series we
can use all of IOMMU_DOMAIN_DMA(_FQ, _SQ) on any PCI device but we want
to default to either IOMMU_DOMAIN_DMA_FQ or IOMMU_DOMAIN_SQ based on
whether we're running in a paging hypervisor (z/VM or KVM) to get the
best performance. From a semantic point of view I felt that this is a
good match for ops->def_domain_type in that we pick a default but it's
still possible to change the domain type e.g. via sysfs. Now this had
the problem that ops->def_domain_type would cause IOMMU_DOMAIN_DMA_FQ
to be used even if iommu_set_dma_strict() was called (via
iommu.strict=1) because there is a undocumented override of ops-
>def_domain_type over iommu_def_domain_type which I believe comes from
the mixing of capability and default you also point at.

I think ideally we need two separate mechanism to determine which
domain types can be used for a particular device (capability) and for
which one to default to with the latter part having a clear precedence
between the options. Put together I think iommu.strict=1 should
override a device's preference (ops->def_domain_type) and
CONFIG_IOMMU_DEFAULT_DMA to use IOMMU_DOMAIN_DMA but of course only if
the device is capable of that. Does that sound reasonable?