Re: [Qemu-devel] [V6 4/4] hw/pci-host: Emulate AMD IOMMU

[Michael S. Tsirkin]

On Sun, Mar 13, 2016 at 03:14:37AM +0300, David Kiarie wrote:
> On Fri, Mar 11, 2016 at 4:22 PM, Michael S. Tsirkin <address@hidden> wrote:
> > On Sun, Feb 21, 2016 at 09:11:00PM +0300, David Kiarie wrote:
> >> Add AMD IOMMU emulation support to q35 chipset
> >>
> >> Signed-off-by: David Kiarie <address@hidden>
> >> ---
> >>  hw/pci-host/piix.c            |  1 +
> >>  hw/pci-host/q35.c             | 14 ++++++++++++--
> >>  include/hw/i386/intel_iommu.h |  1 +
> >>  3 files changed, 14 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/hw/pci-host/piix.c b/hw/pci-host/piix.c
> >> index 41aa66f..ab2e24a 100644
> >> --- a/hw/pci-host/piix.c
> >> +++ b/hw/pci-host/piix.c
> >> @@ -36,6 +36,7 @@
> >>  #include "hw/i386/ioapic.h"
> >>  #include "qapi/visitor.h"
> >>  #include "qemu/error-report.h"
> >> +#include "hw/i386/amd_iommu.h"
> >>
> >>  /*
> >>   * I440FX chipset data sheet.
> >
> > Why is this needed?
> >
> >> diff --git a/hw/pci-host/q35.c b/hw/pci-host/q35.c
> >> index 115fb8c..355fb32 100644
> >> --- a/hw/pci-host/q35.c
> >> +++ b/hw/pci-host/q35.c
> >> @@ -31,6 +31,7 @@
> >>  #include "hw/hw.h"
> >>  #include "hw/pci-host/q35.h"
> >>  #include "qapi/visitor.h"
> >> +#include "hw/i386/amd_iommu.h"
> >>
> >>  
> >> /****************************************************************************
> >>   * Q35 host
> >> @@ -505,9 +506,18 @@ static void mch_realize(PCIDevice *d, Error **errp)
> >>                   mch->pci_address_space, &mch->pam_regions[i+1],
> >>                   PAM_EXPAN_BASE + i * PAM_EXPAN_SIZE, PAM_EXPAN_SIZE);
> >>      }
> >> -    /* Intel IOMMU (VT-d) */
> >> -    if (object_property_get_bool(qdev_get_machine(), "iommu", NULL)) {
> >> +
> >> +    if (g_strcmp0(MACHINE(qdev_get_machine())->iommu, INTEL_IOMMU_STR) == 
> >> 0) {
> >> +        /* Intel IOMMU (VT-d) */
> >>          mch_init_dmar(mch);
> >> +    } else if (g_strcmp0(MACHINE(qdev_get_machine())->iommu, 
> >> AMD_IOMMU_STR)
> >> +               == 0) {
> >> +        AMDIOMMUState *iommu_state;
> >> +        PCIDevice *iommu;
> >> +        PCIBus *bus = PCI_BUS(qdev_get_parent_bus(DEVICE(mch)));
> >> +        iommu = pci_create_simple(bus, 0x20, TYPE_AMD_IOMMU_DEVICE);
> >
> > address can be set through a property.
> 
> I missed something here, what is the problem ?Is it the hardcoded address ?

As long as you use pci_create_simple, it's not a problem.
If eventually we manage to switch to using -device,
it'll be useful that by setting address property
machine can influence address for the device.


> >
> >> +        iommu_state = AMD_IOMMU_DEVICE(iommu);
> >> +        pci_setup_iommu(bus, bridge_host_amd_iommu, iommu_state);
> >
> >
> > It would be better to move this chunk to a separate function.
> >
> >>      }
> >>  }
> >>
> >> diff --git a/include/hw/i386/intel_iommu.h b/include/hw/i386/intel_iommu.h
> >> index b024ffa..539530c 100644
> >> --- a/include/hw/i386/intel_iommu.h
> >> +++ b/include/hw/i386/intel_iommu.h
> >> @@ -27,6 +27,7 @@
> >>  #define TYPE_INTEL_IOMMU_DEVICE "intel-iommu"
> >>  #define INTEL_IOMMU_DEVICE(obj) \
> >>       OBJECT_CHECK(IntelIOMMUState, (obj), TYPE_INTEL_IOMMU_DEVICE)
> >> +#define INTEL_IOMMU_STR "intel"
> >>
> >>  /* DMAR Hardware Unit Definition address (IOMMU unit) */
> >>  #define Q35_HOST_BRIDGE_IOMMU_ADDR  0xfed90000ULL
> >> --
> >> 2.1.4