Re: [PATCH] vhost-vdpa: skip TPM CRB memory section

[Michael S. Tsirkin]

On Wed, Nov 23, 2022 at 12:25:58PM +0100, Philippe Mathieu-Daudé wrote:
> On 23/11/22 08:23, Michael S. Tsirkin wrote:
> > On Wed, Nov 23, 2022 at 11:20:41AM +0400, Marc-André Lureau wrote:
> > > Hi
> > > 
> > > On Wed, Nov 23, 2022 at 12:32 AM Michael S. Tsirkin <mst@redhat.com> 
> > > wrote:
> > > 
> > >      On Tue, Nov 22, 2022 at 06:53:49PM +0400, marcandre.lureau@redhat.com
> > >      wrote:
> > >      > From: Marc-André Lureau <marcandre.lureau@redhat.com>
> > >      >
> > >      > 851d6d1a0f ("vfio/common: remove spurious tpm-crb-cmd misalignment
> > >      > warning") removed the warning on vfio_listener_region_add() path.
> > >      >
> > >      > An error is reported for vhost-vdpa case:
> > >      > qemu-kvm: vhost_vdpa_listener_region_add received unaligned region
> > >      >
> > >      > Skip the CRB device.
> > >      >
> > >      > Fixes:
> > >      > https://bugzilla.redhat.com/show_bug.cgi?id=2141965
> > >      >
> > >      > Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
> > >      > ---
> > >      >  hw/virtio/vhost-vdpa.c | 6 ++++++
> > >      >  1 file changed, 6 insertions(+)
> > >      >
> > >      > diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c
> > >      > index 7468e44b87..9d7206e4b8 100644
> > >      > --- a/hw/virtio/vhost-vdpa.c
> > >      > +++ b/hw/virtio/vhost-vdpa.c
> > >      > @@ -19,6 +19,7 @@
> > >      >  #include "hw/virtio/virtio-net.h"
> > >      >  #include "hw/virtio/vhost-shadow-virtqueue.h"
> > >      >  #include "hw/virtio/vhost-vdpa.h"
> > >      > +#include "sysemu/tpm.h"
> > >      >  #include "exec/address-spaces.h"
> > >      >  #include "migration/blocker.h"
> > >      >  #include "qemu/cutils.h"
> > >      > @@ -46,6 +47,11 @@ static bool vhost_vdpa_listener_skipped_section
> > >      (MemoryRegionSection *section,
> > >      >  {
> > >      >      Int128 llend;
> > >      >
> > >      > +    if (TPM_IS_CRB(section->mr->owner)) {
> > >      > +        /* The CRB command buffer has its base address unaligned. 
> > > */
> > >      > +        return true;
> > >      > +    }
> > >      > +
> > > 
> > >      Quite a hack. We can't really keep adding dependency on random 
> > > devices
> > > 
> > > 
> > > Agree it's not great. but it's not strictly a dependency. At least you can
> > > still build with !CONFIG_TPM.
> > 
> > but what does it have to do with tpm?
> > 
> > 
> > > 
> > >      to vhost. And would you add hacks like this to listeners?
> > >      Pls figure out what's special about this buffer.
> > >      Also if this section is unaligned then doesn't it break up
> > >      other aligned sections?
> > > 
> > > 
> > > See the original discussion:
> > > https://patchew.org/QEMU/20220208133842.112017-1-eric.auger@redhat.com/
> > > 20220208133842.112017-2-eric.auger@redhat.com/
> > > 
> > > It is not clear whether aligning the tpm-crb-cmd region would work 
> > > (overlapping
> > > tpm-crb-mmio).
> > > 
> > > Peter Maydell said: "There's nothing that guarantees alignment for memory
> > > regions at all, whether they're RAM, IO or anything else.".
> > > 
> > > Maybe vfio/vhost should simply skip those odd regions silently.
> > 
> > 
> > How do we detect them? Are these regions not DMA-able on real hardware?
> 
> Are you suggesting expanding the Memory API?
> 
> - Add 'base_address_unaligned' to MemoryRegion structure (default to
>   false)
> 
> - Add memory_region_set_..._unaligned()
> 
> Then have TYPE_TPM_CRB use this API?

Fundamentally since we for some reason don't seem to want to use
existing dma ram api, then yes. But let's have it say what exactly
it means, the fact it's not aligned is not the reason we don't need
to DMA there.


-- 
MST