Quoting David Gibson (2017-02-28 20:56:44)
> On Tue, Feb 28, 2017 at 06:56:47PM -0600, Michael Roth wrote:
> > In certain cases, such as PCI-passthrough with VFIO, we cannot offload
> > MMIO accesses to KVM unless the BAR alignment matches the host. This
> > patch, in conjunction with a separately submitted patch for SLOF
> > which allows for control of this via the device-tree, allows us to
> > set this alignment via QEMU.
> >
> > Cc: qemu-...@nongnu.org
> > Cc: Nikunj A Dadhania <nik...@linux.vnet.ibm.com>
> > Cc: David Gibson <da...@gibson.dropbear.id.au>
> > Cc: Alexey Kardashevskiy <a...@ozlabs.ru>
> > Signed-off-by: Michael Roth <mdr...@linux.vnet.ibm.com>
> > ---
> > hw/ppc/spapr.c | 7 ++++++-
> > hw/ppc/spapr_pci.c | 8 ++++++++
> > include/hw/pci-host/spapr.h | 1 +
> > 3 files changed, 15 insertions(+), 1 deletion(-)
> >
> > diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
> > index 81c6c1c..321951c 100644
> > --- a/hw/ppc/spapr.c
> > +++ b/hw/ppc/spapr.c
> > @@ -3121,7 +3121,12 @@ DEFINE_SPAPR_MACHINE(2_9, "2.9", true);
> > * pseries-2.8
> > */
> > #define SPAPR_COMPAT_2_8 \
> > - HW_COMPAT_2_8
> > + HW_COMPAT_2_8 \
> > + { \
> > + .driver = TYPE_SPAPR_PCI_HOST_BRIDGE, \
> > + .property = "mem_bar_min_align", \
> > + .value = "0", \
> > + }, \
> >
> > static void spapr_machine_2_8_instance_options(MachineState *machine)
> > {
> > diff --git a/hw/ppc/spapr_pci.c b/hw/ppc/spapr_pci.c
> > index 2a3499e..fdfd494 100644
> > --- a/hw/ppc/spapr_pci.c
> > +++ b/hw/ppc/spapr_pci.c
> > @@ -1660,6 +1660,10 @@ static void spapr_phb_realize(DeviceState *dev,
> > Error **errp)
> > return;
> > }
> >
> > + if (sphb->mem_bar_min_align == (uint64_t)-1) {
> > + sphb->mem_bar_min_align = qemu_real_host_page_size;
> > + }
> > +
> > sphb->dtbusname = g_strdup_printf("pci@%" PRIx64, sphb->buid);
> >
> > namebuf = alloca(strlen(sphb->dtbusname) + 32);
> > @@ -1854,6 +1858,8 @@ static Property spapr_phb_properties[] = {
> > DEFINE_PROP_UINT32("numa_node", sPAPRPHBState, numa_node, -1),
> > DEFINE_PROP_BOOL("pre-2.8-migration", sPAPRPHBState,
> > pre_2_8_migration, false),
> > + DEFINE_PROP_UINT64("mem_bar_min_align", sPAPRPHBState,
> > mem_bar_min_align,
> > + -1),
> > DEFINE_PROP_END_OF_LIST(),
> > };
> >
> > @@ -2224,6 +2230,8 @@ int spapr_populate_pci_dt(sPAPRPHBState *phb,
> > if (ret) {
> > return ret;
> > }
> > + _FDT(fdt_setprop_cell(fdt, bus_off, "qemu,mem-bar-min-align",
> > + phb->mem_bar_min_align));
>
> AFAICT the sensible thing to do here is to leave the property out on
> older machine types. In which case you need an if
> (phb->mem_bar_min_align) here.
Yah. On the SLOF side it seemed like 64K was the safer default, which
made this more of an override. But that doesn't make much sense unless
you're mix-and-matching SLOFs/QEMUs. Will change it for v2.
> >
> > return 0;
> > }
> > diff --git a/include/hw/pci-host/spapr.h b/include/hw/pci-host/spapr.h
> > index dfa7614..fa33346 100644
> > --- a/include/hw/pci-host/spapr.h
> > +++ b/include/hw/pci-host/spapr.h
> > @@ -79,6 +79,7 @@ struct sPAPRPHBState {
> > uint64_t dma64_win_addr;
> >
> > uint32_t numa_node;
> > + uint64_t mem_bar_min_align;
> >
> > /* Fields for migration compatibility hacks */
> > bool pre_2_8_migration;
>
> --
> David Gibson | I'll have my music baroque, and my code
> david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_
> | _way_ _around_!
> http://www.ozlabs.org/~dgibson
