Message-ID: <CAD=FV=Vg7kqhgxZppHXwMPMc0xATZ+MqbrXx-FB0eg7pHhNE8w@mail.gmail.com>
Date:   Tue, 22 Jun 2021 09:06:02 -0700
From:   Doug Anderson <dianders@...omium.org>
To:     Robin Murphy <robin.murphy@....com>
Cc:     Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        "Rafael J. Wysocki" <rafael@...nel.org>,
        "Rafael J. Wysocki" <rafael.j.wysocki@...el.com>,
        Will Deacon <will@...nel.org>, Joerg Roedel <joro@...tes.org>,
        Bjorn Andersson <bjorn.andersson@...aro.org>,
        Ulf Hansson <ulf.hansson@...aro.org>,
        Adrian Hunter <adrian.hunter@...el.com>,
        Bjorn Helgaas <bhelgaas@...gle.com>,
        Rob Clark <robdclark@...omium.org>,
        linux-arm-msm <linux-arm-msm@...r.kernel.org>,
        linux-pci@...r.kernel.org, quic_c_gdjako@...cinc.com,
        "list@....net:IOMMU DRIVERS <iommu@...ts.linux-foundation.org>, Joerg
        Roedel <joro@...tes.org>," <iommu@...ts.linux-foundation.org>,
        Sonny Rao <sonnyrao@...omium.org>,
        Sai Prakash Ranjan <saiprakash.ranjan@...eaurora.org>,
        Linux MMC List <linux-mmc@...r.kernel.org>,
        Veerabhadrarao Badiganti <vbadigan@...eaurora.org>,
        Rajat Jain <rajatja@...gle.com>,
        Saravana Kannan <saravanak@...gle.com>,
        Joel Fernandes <joel@...lfernandes.org>,
        Andy Gross <agross@...nel.org>,
        Bartosz Golaszewski <bgolaszewski@...libre.com>,
        Dan Williams <dan.j.williams@...el.com>,
        Geert Uytterhoeven <geert@...ux-m68k.org>,
        Heikki Krogerus <heikki.krogerus@...ux.intel.com>,
        Randy Dunlap <rdunlap@...radead.org>,
        LKML <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH 0/6] iommu: Enable devices to request non-strict DMA,
 starting with QCom SD/MMC

Hi,

On Tue, Jun 22, 2021 at 4:35 AM Robin Murphy <robin.murphy@....com> wrote:
>
> Hi Doug,
>
> On 2021-06-22 00:52, Douglas Anderson wrote:
> >
> > This patch attempts to put forward a proposal for enabling non-strict
> > DMA on a device-by-device basis. The patch series requests non-strict
> > DMA for the Qualcomm SDHCI controller as a first device to enable,
> > getting a nice bump in performance with what's believed to be a very
> > small drop in security / safety (see the patch for the full argument).
> >
> > As part of this patch series I am end up slightly cleaning up some of
> > the interactions between the PCI subsystem and the IOMMU subsystem but
> > I don't go all the way to fully remove all the tentacles. Specifically
> > this patch series only concerns itself with a single aspect: strict
> > vs. non-strict mode for the IOMMU. I'm hoping that this will be easier
> > to talk about / reason about for more subsystems compared to overall
> > deciding what it means for a device to be "external" or "untrusted".
> >
> > If something like this patch series ends up being landable, it will
> > undoubtedly need coordination between many maintainers to land. I
> > believe it's fully bisectable but later patches in the series
> > definitely depend on earlier ones. Sorry for the long CC list. :(
>
> Unfortunately, this doesn't work. In normal operation, the default
> domains should be established long before individual drivers are even
> loaded (if they are modules), let alone anywhere near probing. The fact
> that iommu_probe_device() sometimes gets called far too late off the
> back of driver probe is an unfortunate artefact of the original
> probe-deferral scheme, and causes other problems like potentially
> malformed groups - I've been forming a plan to fix that for a while now,
> so I for one really can't condone anything trying to rely on it.
> Non-deterministic behaviour based on driver probe order for multi-device
> groups is part of the existing problem, and your proposal seems equally
> vulnerable to that too.

Doh! :( I definitely can't say I understand the iommu subsystem
amazingly well. It was working for me, but I could believe that I was
somehow violating a rule somewhere.

I'm having a bit of a hard time understanding where the problem is
though. Is there any chance that you missed the part of my series
where I introduced a "pre_probe" step? Specifically, I see this:

* really_probe() is called w/ a driver and a device.
* -> calls dev->bus->dma_configure() w/ a "struct device *"
* -> eventually calls iommu_probe_device() w/ the device.
* -> calls iommu_alloc_default_domain() w/ the device
* -> calls iommu_group_alloc_default_domain()
* -> always allocates a new domain

...so we always have a "struct device" when a domain is allocated if
that domain is going to be associated with a device.

I will agree that iommu_probe_device() is called before the driver
probe, but unless I missed something it's after the device driver is
loaded.  ...and assuming something like patch #1 in this series looks
OK then iommu_probe_device() will be called after "pre_probe".

So assuming I'm not missing something, I'm not actually relying the
IOMMU getting init off the back of driver probe.


> FWIW we already have a go-faster knob for people who want to tweak the
> security/performance compromise for specific devices, namely the sysfs
> interface for changing a group's domain type before binding the relevant
> driver(s). Is that something you could use in your application, say from
> an initramfs script?

We've never had an initramfs script in Chrome OS. I don't know all the
history of why (I'm trying to check), but I'm nearly certain it was a
conscious decision. Probably it has to do with the fact that we're not
trying to build a generic distribution where a single boot source can
boot a huge variety of hardware. We generally have one kernel for a
class of devices. I believe avoiding the initramfs just keeps things
simpler.

I think trying to revamp Chrome OS to switch to an initramfs type
system would be a pretty big undertaking since (as I understand it)
you can't just run a little command and then return to the normal boot
flow. Once you switch to initramfs you're committing to finding /
setting up the rootfs yourself and on Chrome OS I believe that means a
whole bunch of dm-verity work.


...so probably the initramfs is a no-go for me, but I'm still crossing
my fingers that the pre_probe() might be legit if you take a second
look at it?

-Doug

Powered by blists - more mailing lists