RE: [PATCH v1 07/22] intel_iommu: add set/unset_iommu_context callback

["Liu, Yi L" <yi.l.liu@intel.com>]

> From: Peter Xu <peterx@redhat.com>
> Sent: Tuesday, March 24, 2020 5:29 AM
> To: Liu, Yi L <yi.l.liu@intel.com>
> Subject: Re: [PATCH v1 07/22] intel_iommu: add set/unset_iommu_context
> callback
> 
> On Sun, Mar 22, 2020 at 05:36:04AM -0700, Liu Yi L wrote:
> > This patch adds set/unset_iommu_context() impelementation in Intel
> > vIOMMU. For Intel platform, pass-through modules (e.g. VFIO) could
> > set HostIOMMUContext to Intel vIOMMU emulator.
> >
> > Cc: Kevin Tian <kevin.tian@intel.com>
> > Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
> > Cc: Peter Xu <peterx@redhat.com>
> > Cc: Yi Sun <yi.y.sun@linux.intel.com>
> > Cc: Paolo Bonzini <pbonzini@redhat.com>
> > Cc: Richard Henderson <rth@twiddle.net>
> > Cc: Eduardo Habkost <ehabkost@redhat.com>
> > Signed-off-by: Liu Yi L <yi.l.liu@intel.com>
> > ---
> >  hw/i386/intel_iommu.c         | 70
> +++++++++++++++++++++++++++++++++++++++----
> >  include/hw/i386/intel_iommu.h | 17 +++++++++--
> >  2 files changed, 80 insertions(+), 7 deletions(-)
> >
> > diff --git a/hw/i386/intel_iommu.c b/hw/i386/intel_iommu.c
> > index 4b22910..8d9204f 100644
> > --- a/hw/i386/intel_iommu.c
> > +++ b/hw/i386/intel_iommu.c
> > @@ -3354,23 +3354,35 @@ static const MemoryRegionOps vtd_mem_ir_ops =
> {
> >      },
> >  };
> >
> > -VTDAddressSpace *vtd_find_add_as(IntelIOMMUState *s, PCIBus *bus, int
> devfn)
> > +/**
> > + * Fetch a VTDBus instance for given PCIBus. If no existing instance,
> > + * allocate one.
> > + */
> > +static VTDBus *vtd_find_add_bus(IntelIOMMUState *s, PCIBus *bus)
> >  {
> >      uintptr_t key = (uintptr_t)bus;
> >      VTDBus *vtd_bus = g_hash_table_lookup(s->vtd_as_by_busptr, &key);
> > -    VTDAddressSpace *vtd_dev_as;
> > -    char name[128];
> >
> >      if (!vtd_bus) {
> >          uintptr_t *new_key = g_malloc(sizeof(*new_key));
> >          *new_key = (uintptr_t)bus;
> >          /* No corresponding free() */
> > -        vtd_bus = g_malloc0(sizeof(VTDBus) + sizeof(VTDAddressSpace *) * \
> > -                            PCI_DEVFN_MAX);
> > +        vtd_bus = g_malloc0(sizeof(VTDBus) + PCI_DEVFN_MAX * \
> > +                            (sizeof(VTDAddressSpace *) + \
> > +                             sizeof(VTDHostIOMMUContext *)));
> 
> IIRC I commented on this before...  Shouldn't sizeof(VTDBus) be
> enough?

Right. My bad. Will do it in next version.

> >          vtd_bus->bus = bus;
> >          g_hash_table_insert(s->vtd_as_by_busptr, new_key, vtd_bus);
> >      }
> > +    return vtd_bus;
> > +}
> > +
> > +VTDAddressSpace *vtd_find_add_as(IntelIOMMUState *s, PCIBus *bus, int
> devfn)
> > +{
> > +    VTDBus *vtd_bus;
> > +    VTDAddressSpace *vtd_dev_as;
> > +    char name[128];
> >
> > +    vtd_bus = vtd_find_add_bus(s, bus);
> >      vtd_dev_as = vtd_bus->dev_as[devfn];
> >
> >      if (!vtd_dev_as) {
> > @@ -3436,6 +3448,52 @@ VTDAddressSpace
> *vtd_find_add_as(IntelIOMMUState *s, PCIBus *bus, int devfn)
> >      return vtd_dev_as;
> >  }
> >
> > +static int vtd_dev_set_iommu_context(PCIBus *bus, void *opaque,
> > +                                     int devfn,
> > +                                     HostIOMMUContext *host_icx)
> > +{
> > +    IntelIOMMUState *s = opaque;
> > +    VTDBus *vtd_bus;
> > +    VTDHostIOMMUContext *vtd_dev_icx;
> > +
> > +    assert(0 <= devfn && devfn < PCI_DEVFN_MAX);
> > +
> > +    vtd_bus = vtd_find_add_bus(s, bus);
> > +
> > +    vtd_iommu_lock(s);
> > +    vtd_dev_icx = vtd_bus->dev_icx[devfn];
> > +
> > +    if (!vtd_dev_icx) {
> 
> We can assert this directly I think, in case we accidentally set the
> context twice without notice.

good idea. will add it.

> 
> > +        vtd_bus->dev_icx[devfn] = vtd_dev_icx =
> > +                    g_malloc0(sizeof(VTDHostIOMMUContext));
> > +        vtd_dev_icx->vtd_bus = vtd_bus;
> > +        vtd_dev_icx->devfn = (uint8_t)devfn;
> > +        vtd_dev_icx->iommu_state = s;
> > +        vtd_dev_icx->host_icx = host_icx;
> > +    }
> > +    vtd_iommu_unlock(s);
> > +
> > +    return 0;
> > +}
> > +
> > +static void vtd_dev_unset_iommu_context(PCIBus *bus, void *opaque, int
> devfn)
> > +{
> > +    IntelIOMMUState *s = opaque;
> > +    VTDBus *vtd_bus;
> > +    VTDHostIOMMUContext *vtd_dev_icx;
> > +
> > +    assert(0 <= devfn && devfn < PCI_DEVFN_MAX);
> > +
> > +    vtd_bus = vtd_find_add_bus(s, bus);
> > +
> > +    vtd_iommu_lock(s);
> > +
> > +    vtd_dev_icx = vtd_bus->dev_icx[devfn];
> > +    g_free(vtd_dev_icx);
> 
> Better set it as NULL, and can also drop vtd_dev_icx which seems
> meaningless..

right. it's missed.

>        g_free(vtd_bus->dev_icx[devfn]);
>        vtd_bus->dev_icx[devfn] = NULL;
> 
> > +
> > +    vtd_iommu_unlock(s);
> > +}
> > +
> >  static uint64_t get_naturally_aligned_size(uint64_t start,
> >                                             uint64_t size, int gaw)
> >  {
> > @@ -3731,6 +3789,8 @@ static AddressSpace *vtd_host_dma_iommu(PCIBus
> *bus, void *opaque, int devfn)
> >
> >  static PCIIOMMUOps vtd_iommu_ops = {
> >      .get_address_space = vtd_host_dma_iommu,
> > +    .set_iommu_context = vtd_dev_set_iommu_context,
> > +    .unset_iommu_context = vtd_dev_unset_iommu_context,
> >  };
> >
> >  static bool vtd_decide_config(IntelIOMMUState *s, Error **errp)
> > diff --git a/include/hw/i386/intel_iommu.h b/include/hw/i386/intel_iommu.h
> > index 3870052..9b4fc0a 100644
> > --- a/include/hw/i386/intel_iommu.h
> > +++ b/include/hw/i386/intel_iommu.h
> > @@ -64,6 +64,7 @@ typedef union VTD_IR_TableEntry VTD_IR_TableEntry;
> >  typedef union VTD_IR_MSIAddress VTD_IR_MSIAddress;
> >  typedef struct VTDPASIDDirEntry VTDPASIDDirEntry;
> >  typedef struct VTDPASIDEntry VTDPASIDEntry;
> > +typedef struct VTDHostIOMMUContext VTDHostIOMMUContext;
> >
> >  /* Context-Entry */
> >  struct VTDContextEntry {
> > @@ -112,10 +113,20 @@ struct VTDAddressSpace {
> >      IOVATree *iova_tree;          /* Traces mapped IOVA ranges */
> >  };
> >
> > +struct VTDHostIOMMUContext {
> > +    VTDBus *vtd_bus;
> > +    uint8_t devfn;
> > +    HostIOMMUContext *host_icx;
> > +    IntelIOMMUState *iommu_state;
> > +};
> > +
> >  struct VTDBus {
> > -    PCIBus* bus;		/* A reference to the bus to provide translation
> for */
> > +    /* A reference to the bus to provide translation for */
> > +    PCIBus *bus;
> >      /* A table of VTDAddressSpace objects indexed by devfn */
> > -    VTDAddressSpace *dev_as[];
> > +    VTDAddressSpace *dev_as[PCI_DEVFN_MAX];
> > +    /* A table of VTDHostIOMMUContext objects indexed by devfn */
> > +    VTDHostIOMMUContext *dev_icx[PCI_DEVFN_MAX];
> >  };
> >
> >  struct VTDIOTLBEntry {
> > @@ -271,6 +282,8 @@ struct IntelIOMMUState {
> >      /*
> >       * Protects IOMMU states in general.  Currently it protects the
> >       * per-IOMMU IOTLB cache, and context entry cache in VTDAddressSpace.
> > +     * Protect the update/usage of HostIOMMUContext pointer cached in
> > +     * VTDBus->dev_icx array as array elements may be updated by hotplug
> 
> I think the context update does not need to be updated, because they
> should always be with the BQL, right?

Hmmmm, maybe I used bad description. My purpose is to protect the stored
HostIOMMUContext pointer in vIOMMU. With pci_device_set/unset_iommu_context,
vIOMMU have a copy of HostIOMMUContext. If VFIO container is released
(e.g. hotpulg out device), HostIOMMUContext will alos be released. This
will trigger the pci_device_unset_iommu_context() to clean the copy. To
avoid using a staled HostIOMMUContext in vIOMMU, vIOMMU should have a
lock to block the pci_device_unset_iommu_context() calling until other
threads finished their HostIOMMUContext usage. Do you want a description
update here or other preference?

Regards,
Yi Liu