Linux Virtualization - Mar 2019 - [RFC PATCH] virtio_ring: Use DMA API if guest memory is encrypted

Hello Michael,

Sorry for the delay in responding. We had some internal discussions on
this.

Michael S. Tsirkin <mst at redhat.com> writes:
> On Mon, Feb 04, 2019 at 04:14:20PM -0200, Thiago Jung Bauermann wrote:
>>
>> Hello Michael,
>>
>> Michael S. Tsirkin <mst at redhat.com> writes:
>>
>> > On Tue, Jan 29, 2019 at 03:42:44PM -0200, Thiago Jung Bauermann
wrote:
>> So while ACCESS_PLATFORM solves our problems for secure guests, we
can't
>> turn it on by default because we can't affect legacy systems. Doing
so
>> would penalize existing systems that can access all memory. They would
>> all have to unnecessarily go through address translations, and take a
>> performance hit.
>
> So as step one, you just give hypervisor admin an option to run legacy
> systems faster by blocking secure mode. I don't see why that is
> so terrible.
There are a few reasons why:

1. It's bad user experience to require people to fiddle with knobs for
obscure reasons if it's possible to design things such that they Just
Work.

2. "User" in this case can be a human directly calling QEMU, but could
also be libvirt or one of its users, or some other framework. This means
having to adjust and/or educate an open-ended number of people and
software. It's best avoided if possible.

3. The hypervisor admin and the admin of the guest system don't
necessarily belong to the same organization (e.g., cloud provider and
cloud customer), so there may be some friction when they need to
coordinate to get this right.

4. A feature of our design is that the guest may or may not decide to
"go secure" at boot time, so it's best not to depend on flags that
may
or may not have been set at the time QEMU was started.
>> The semantics of ACCESS_PLATFORM assume that the hypervisor/QEMU knows
>> in advance - right when the VM is instantiated - that it will not have
>> access to all guest memory.
>
> Not quite. It just means that hypervisor can live with not having
> access to all memory. If platform wants to give it access
> to all memory that is quite all right.
Except that on powerpc it also means "there's an IOMMU present"
and
there's no way to say "bypass IOMMU translation". :-/
>> Another way of looking at this issue which also explains our reluctance
>> is that the only difference between a secure guest and a regular guest
>> (at least regarding virtio) is that the former uses swiotlb while the
>> latter doens't.
>
> But swiotlb is just one implementation. It's a guest internal thing.
The
> issue is that memory isn't host accessible.
>From what I understand of the ACCESS_PLATFORM definition, the host will
only ever try to access memory addresses that are supplied to it by the
guest, so all of the secure guest memory that the host cares about is
accessible:

    If this feature bit is set to 0, then the device has same access to
    memory addresses supplied to it as the driver has. In particular,
    the device will always use physical addresses matching addresses
    used by the driver (typically meaning physical addresses used by the
    CPU) and not translated further, and can access any address supplied
    to it by the driver. When clear, this overrides any
    platform-specific description of whether device access is limited or
    translated in any way, e.g. whether an IOMMU may be present.

All of the above is true for POWER guests, whether they are secure
guests or not.

Or are you saying that a virtio device may want to access memory
addresses that weren't supplied to it by the driver?
>> And from the device's point of view they're
>> indistinguishable. It can't tell one guest that is using swiotlb
from
>> one that isn't. And that implies that secure guest vs regular guest
>> isn't a virtio interface issue, it's "guest internal
affairs". So
>> there's no reason to reflect that in the feature flags.
>
> So don't. The way not to reflect that in the feature flags is
> to set ACCESS_PLATFORM.  Then you say *I don't care let platform
device*.
>
>
> Without ACCESS_PLATFORM
> virtio has a very specific opinion about the security of the
> device, and that opinion is that device is part of the guest
> supervisor security domain.
Sorry for being a bit dense, but not sure what "the device is part of
the guest supervisor security domain" means. In powerpc-speak,
"supervisor" is the operating system so perhaps that explains my
confusion. Are you saying that without ACCESS_PLATFORM, the guest
considers the host to be part of the guest operating system's security
domain? If so, does that have any other implication besides "the host
can access any address supplied to it by the driver"? If that is the
case, perhaps the definition of ACCESS_PLATFORM needs to be amended to
include that information because it's not part of the current
definition.
>> That said, we still would like to arrive at a proper design for this
>> rather than add yet another hack if we can avoid it. So here's
another
>> proposal: considering that the dma-direct code (in kernel/dma/direct.c)
>> automatically uses swiotlb when necessary (thanks to Christoph's
recent
>> DMA work), would it be ok to replace virtio's own direct-memory
code
>> that is used in the !ACCESS_PLATFORM case with the dma-direct code?
That
>> way we'll get swiotlb even with !ACCESS_PLATFORM, and virtio will
get a
>> code cleanup (replace open-coded stuff with calls to existing
>> infrastructure).
>
> Let's say I have some doubts that there's an API that
> matches what virtio with its bag of legacy compatibility exactly.
Ok.
>> > But the name "sev_active" makes me scared because at
least AMD guys who
>> > were doing the sensible thing and setting ACCESS_PLATFORM
>>
>> My understanding is, AMD guest-platform knows in advance that their
>> guest will run in secure mode and hence sets the flag at the time of VM
>> instantiation. Unfortunately we dont have that luxury on our platforms.
>
> Well you do have that luxury. It looks like that there are existing
> guests that already acknowledge ACCESS_PLATFORM and you are not happy
> with how that path is slow. So you are trying to optimize for
> them by clearing ACCESS_PLATFORM and then you have lost ability
> to invoke DMA API.
>
> For example if there was another flag just like ACCESS_PLATFORM
> just not yet used by anyone, you would be all fine using that right?
Yes, a new flag sounds like a great idea. What about the definition
below?

VIRTIO_F_ACCESS_PLATFORM_NO_IOMMU This feature has the same meaning as
    VIRTIO_F_ACCESS_PLATFORM both when set and when not set, with the
    exception that the IOMMU is explicitly defined to be off or bypassed
    when accessing memory addresses supplied to the device by the
    driver. This flag should be set by the guest if offered, but to
    allow for backward-compatibility device implementations allow for it
    to be left unset by the guest. It is an error to set both this flag
    and VIRTIO_F_ACCESS_PLATFORM.
> Is there any justification to doing that beyond someone putting
> out slow code in the past?
The definition of the ACCESS_PLATFORM flag is generic and captures the
notion of memory access restrictions for the device. Unfortunately, on
powerpc pSeries guests it also implies that the IOMMU is turned on even
though pSeries guests have never used IOMMU for virtio devices. Combined
with the lack of a way to turn off or bypass the IOMMU for virtio
devices, this means that existing guests in the field are compelled to
use the IOMMU even though that never was the case before, and said
guests having no mechanism to turn it off.

Therefore, we need a new flag to signal the memory access restriction
present in secure guests which doesn't also imply turning on the IOMMU.

--
Thiago Jung Bauermann
IBM Linux Technology Center

On Wed, Mar 20, 2019 at 01:13:41PM -0300, Thiago Jung Bauermann
wrote:
> >> Another way of looking at this issue which also explains our
reluctance
> >> is that the only difference between a secure guest and a regular
guest
> >> (at least regarding virtio) is that the former uses swiotlb while
the
> >> latter doens't.
> >
> > But swiotlb is just one implementation. It's a guest internal
thing. The
> > issue is that memory isn't host accessible.
> 
> >From what I understand of the ACCESS_PLATFORM definition, the host will
> only ever try to access memory addresses that are supplied to it by the
> guest, so all of the secure guest memory that the host cares about is
> accessible:
> 
>     If this feature bit is set to 0, then the device has same access to
>     memory addresses supplied to it as the driver has. In particular,
>     the device will always use physical addresses matching addresses
>     used by the driver (typically meaning physical addresses used by the
>     CPU) and not translated further, and can access any address supplied
>     to it by the driver. When clear, this overrides any
>     platform-specific description of whether device access is limited or
>     translated in any way, e.g. whether an IOMMU may be present.
> 
> All of the above is true for POWER guests, whether they are secure
> guests or not.
> 
> Or are you saying that a virtio device may want to access memory
> addresses that weren't supplied to it by the driver?
Your logic would apply to IOMMUs as well.  For your mode, there are
specific encrypted memory regions that driver has access to but device
does not. that seems to violate the constraint.

> >> And from the device's point of view they're
> >> indistinguishable. It can't tell one guest that is using
swiotlb from
> >> one that isn't. And that implies that secure guest vs regular
guest
> >> isn't a virtio interface issue, it's "guest internal
affairs". So
> >> there's no reason to reflect that in the feature flags.
> >
> > So don't. The way not to reflect that in the feature flags is
> > to set ACCESS_PLATFORM.  Then you say *I don't care let platform
device*.
> >
> >
> > Without ACCESS_PLATFORM
> > virtio has a very specific opinion about the security of the
> > device, and that opinion is that device is part of the guest
> > supervisor security domain.
> 
> Sorry for being a bit dense, but not sure what "the device is part of
> the guest supervisor security domain" means. In powerpc-speak,
> "supervisor" is the operating system so perhaps that explains my
> confusion. Are you saying that without ACCESS_PLATFORM, the guest
> considers the host to be part of the guest operating system's security
> domain?
I think so. The spec says "device has same access as driver".
> If so, does that have any other implication besides "the host
> can access any address supplied to it by the driver"? If that is the
> case, perhaps the definition of ACCESS_PLATFORM needs to be amended to
> include that information because it's not part of the current
> definition.
> 
> >> That said, we still would like to arrive at a proper design for
this
> >> rather than add yet another hack if we can avoid it. So here's
another
> >> proposal: considering that the dma-direct code (in
kernel/dma/direct.c)
> >> automatically uses swiotlb when necessary (thanks to
Christoph's recent
> >> DMA work), would it be ok to replace virtio's own
direct-memory code
> >> that is used in the !ACCESS_PLATFORM case with the dma-direct
code? That
> >> way we'll get swiotlb even with !ACCESS_PLATFORM, and virtio
will get a
> >> code cleanup (replace open-coded stuff with calls to existing
> >> infrastructure).
> >
> > Let's say I have some doubts that there's an API that
> > matches what virtio with its bag of legacy compatibility exactly.
> 
> Ok.
> 
> >> > But the name "sev_active" makes me scared because
at least AMD guys who
> >> > were doing the sensible thing and setting ACCESS_PLATFORM
> >>
> >> My understanding is, AMD guest-platform knows in advance that
their
> >> guest will run in secure mode and hence sets the flag at the time
of VM
> >> instantiation. Unfortunately we dont have that luxury on our
platforms.
> >
> > Well you do have that luxury. It looks like that there are existing
> > guests that already acknowledge ACCESS_PLATFORM and you are not happy
> > with how that path is slow. So you are trying to optimize for
> > them by clearing ACCESS_PLATFORM and then you have lost ability
> > to invoke DMA API.
> >
> > For example if there was another flag just like ACCESS_PLATFORM
> > just not yet used by anyone, you would be all fine using that right?
> 
> Yes, a new flag sounds like a great idea. What about the definition
> below?
> 
> VIRTIO_F_ACCESS_PLATFORM_NO_IOMMU This feature has the same meaning as
>     VIRTIO_F_ACCESS_PLATFORM both when set and when not set, with the
>     exception that the IOMMU is explicitly defined to be off or bypassed
>     when accessing memory addresses supplied to the device by the
>     driver. This flag should be set by the guest if offered, but to
>     allow for backward-compatibility device implementations allow for it
>     to be left unset by the guest. It is an error to set both this flag
>     and VIRTIO_F_ACCESS_PLATFORM.
It looks kind of narrow but it's an option.
I wonder how we'll define what's an iommu though.

Another idea is maybe something like virtio-iommu?
> > Is there any justification to doing that beyond someone putting
> > out slow code in the past?
> 
> The definition of the ACCESS_PLATFORM flag is generic and captures the
> notion of memory access restrictions for the device. Unfortunately, on
> powerpc pSeries guests it also implies that the IOMMU is turned on
IIUC that's really because on pSeries IOMMU is *always* turned on.
Platform has no way to say what you want it to say
which is bypass the iommu for the specific device.
> even
> though pSeries guests have never used IOMMU for virtio devices. Combined
> with the lack of a way to turn off or bypass the IOMMU for virtio
> devices, this means that existing guests in the field are compelled to
> use the IOMMU even though that never was the case before, and said
> guests having no mechanism to turn it off.
> 
> Therefore, we need a new flag to signal the memory access restriction
> present in secure guests which doesn't also imply turning on the IOMMU.
> 
> --
> Thiago Jung Bauermann
> IBM Linux Technology Center

Michael S. Tsirkin <mst at redhat.com> writes:
> On Wed, Mar 20, 2019 at 01:13:41PM -0300, Thiago Jung Bauermann wrote:
>> >> Another way of looking at this issue which also explains our
reluctance
>> >> is that the only difference between a secure guest and a
regular guest
>> >> (at least regarding virtio) is that the former uses swiotlb
while the
>> >> latter doens't.
>> >
>> > But swiotlb is just one implementation. It's a guest internal
thing. The
>> > issue is that memory isn't host accessible.
>>
>> >From what I understand of the ACCESS_PLATFORM definition, the host
will
>> only ever try to access memory addresses that are supplied to it by the
>> guest, so all of the secure guest memory that the host cares about is
>> accessible:
>>
>>     If this feature bit is set to 0, then the device has same access to
>>     memory addresses supplied to it as the driver has. In particular,
>>     the device will always use physical addresses matching addresses
>>     used by the driver (typically meaning physical addresses used by
the
>>     CPU) and not translated further, and can access any address
supplied
>>     to it by the driver. When clear, this overrides any
>>     platform-specific description of whether device access is limited
or
>>     translated in any way, e.g. whether an IOMMU may be present.
>>
>> All of the above is true for POWER guests, whether they are secure
>> guests or not.
>>
>> Or are you saying that a virtio device may want to access memory
>> addresses that weren't supplied to it by the driver?
>
> Your logic would apply to IOMMUs as well.  For your mode, there are
> specific encrypted memory regions that driver has access to but device
> does not. that seems to violate the constraint.
Right, if there's a pre-configured 1:1 mapping in the IOMMU such that
the device can ignore the IOMMU for all practical purposes I would
indeed say that the logic would apply to IOMMUs as well. :-)

I guess I'm still struggling with the purpose of signalling to the
driver that the host may not have access to memory addresses that it
will never try to access.
>> >> And from the device's point of view they're
>> >> indistinguishable. It can't tell one guest that is using
swiotlb from
>> >> one that isn't. And that implies that secure guest vs
regular guest
>> >> isn't a virtio interface issue, it's "guest
internal affairs". So
>> >> there's no reason to reflect that in the feature flags.
>> >
>> > So don't. The way not to reflect that in the feature flags is
>> > to set ACCESS_PLATFORM.  Then you say *I don't care let
platform device*.
>> >
>> >
>> > Without ACCESS_PLATFORM
>> > virtio has a very specific opinion about the security of the
>> > device, and that opinion is that device is part of the guest
>> > supervisor security domain.
>>
>> Sorry for being a bit dense, but not sure what "the device is part
of
>> the guest supervisor security domain" means. In powerpc-speak,
>> "supervisor" is the operating system so perhaps that explains
my
>> confusion. Are you saying that without ACCESS_PLATFORM, the guest
>> considers the host to be part of the guest operating system's
security
>> domain?
>
> I think so. The spec says "device has same access as driver".
Ok, makes sense.
>> If so, does that have any other implication besides "the host
>> can access any address supplied to it by the driver"? If that is
the
>> case, perhaps the definition of ACCESS_PLATFORM needs to be amended to
>> include that information because it's not part of the current
>> definition.
>>
>> >> > But the name "sev_active" makes me scared
because at least AMD guys who
>> >> > were doing the sensible thing and setting ACCESS_PLATFORM
>> >>
>> >> My understanding is, AMD guest-platform knows in advance that
their
>> >> guest will run in secure mode and hence sets the flag at the
time of VM
>> >> instantiation. Unfortunately we dont have that luxury on our
platforms.
>> >
>> > Well you do have that luxury. It looks like that there are
existing
>> > guests that already acknowledge ACCESS_PLATFORM and you are not
happy
>> > with how that path is slow. So you are trying to optimize for
>> > them by clearing ACCESS_PLATFORM and then you have lost ability
>> > to invoke DMA API.
>> >
>> > For example if there was another flag just like ACCESS_PLATFORM
>> > just not yet used by anyone, you would be all fine using that
right?
>>
>> Yes, a new flag sounds like a great idea. What about the definition
>> below?
>>
>> VIRTIO_F_ACCESS_PLATFORM_NO_IOMMU This feature has the same meaning as
>>     VIRTIO_F_ACCESS_PLATFORM both when set and when not set, with the
>>     exception that the IOMMU is explicitly defined to be off or
bypassed
>>     when accessing memory addresses supplied to the device by the
>>     driver. This flag should be set by the guest if offered, but to
>>     allow for backward-compatibility device implementations allow for
it
>>     to be left unset by the guest. It is an error to set both this flag
>>     and VIRTIO_F_ACCESS_PLATFORM.
>
> It looks kind of narrow but it's an option.
Great!
> I wonder how we'll define what's an iommu though.
Hm, it didn't occur to me it could be an issue. I'll try.
> Another idea is maybe something like virtio-iommu?
You mean, have legacy guests use virtio-iommu to request an IOMMU
bypass? If so, it's an interesting idea for new guests but it doesn't
help with guests that are out today in the field, which don't have A
virtio-iommu driver.
>> > Is there any justification to doing that beyond someone putting
>> > out slow code in the past?
>>
>> The definition of the ACCESS_PLATFORM flag is generic and captures the
>> notion of memory access restrictions for the device. Unfortunately, on
>> powerpc pSeries guests it also implies that the IOMMU is turned on
>
> IIUC that's really because on pSeries IOMMU is *always* turned on.
> Platform has no way to say what you want it to say
> which is bypass the iommu for the specific device.
Yes, that's correct. pSeries guests running on KVM are in a gray area
where theoretically they use an IOMMU but in practice KVM ignores it.
It's unfortunate but it's the reality on the ground today. :-/

--
Thiago Jung Bauermann
IBM Linux Technology Center