On 12/04/2017 08:52 AM, David Gibson wrote:
> On Thu, Nov 23, 2017 at 02:29:51PM +0100, Cédric Le Goater wrote:
>> When the XIVE interrupt mode is activated, the machine needs to expose
>> to the guest the MMIO regions use by the controller :
>>
>> - Event State Buffer (ESB)
>> - Thread Interrupt Management Area (TIMA)
>>
>> Migration will also need to reflect the current interrupt mode in use.
>>
>> Signed-off-by: Cédric Le Goater <c...@kaod.org>
>> ---
>> hw/intc/spapr_xive_hcall.c | 14 ++++++++++++++
>> hw/ppc/spapr.c | 5 +++++
>> include/hw/ppc/spapr_xive.h | 1 +
>> 3 files changed, 20 insertions(+)
>>
>> diff --git a/hw/intc/spapr_xive_hcall.c b/hw/intc/spapr_xive_hcall.c
>> index 60c6c9f4be8f..ba217144878e 100644
>> --- a/hw/intc/spapr_xive_hcall.c
>> +++ b/hw/intc/spapr_xive_hcall.c
>> @@ -933,3 +933,17 @@ void spapr_xive_populate(sPAPRMachineState *spapr, int
>> nr_servers,
>> _FDT(fdt_setprop(fdt, 0, "ibm,plat-res-int-priorities",
>> plat_res_int_priorities,
>> sizeof(plat_res_int_priorities)));
>> }
>> +
>> +void spapr_xive_mmio_map(sPAPRMachineState *spapr)
>> +{
>> + sPAPRXive *xive = spapr->xive;
>> +
>> + /* ESBs */
>> + sysbus_mmio_map(SYS_BUS_DEVICE(xive), 0, xive->esb_base);
>> +
>> + /* Thread Management Interrupt Areas */
>> + /* TODO: Only map the OS TIMA for the moment. Mapping the whole
>> + * region needs some rework in the handlers */
>> + sysbus_mmio_map(SYS_BUS_DEVICE(xive), 1,
>> + xive->tm_base + (1 << xive->tm_shift));
>
> You probably shouldn't be exposing the user TIMA in the DT if you're
> only allowing the OS TIME to be mapped.
The specs requires to map both Uset and OS TIMA.
>
>> +}
>> diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
>> index 3a62369883cc..734706c18cb3 100644
>> --- a/hw/ppc/spapr.c
>> +++ b/hw/ppc/spapr.c
>> @@ -1132,6 +1132,7 @@ static void *spapr_build_fdt(sPAPRMachineState *spapr,
>> } else {
>> /* Populate device tree for XIVE */
>> spapr_xive_populate(spapr, xics_max_server_number(), fdt,
>> PHANDLE_XICP);
>> + spapr_xive_mmio_map(spapr);
>
> This doesn't belong here, spapr_build_fdt() should _just_ build the
> fdt, not have side effects on the actual device state.
Yes. I will move the rest of the XIVE setup in the reset handler
before the device tree is built.
Thanks,
C.
>> }
>>
>> ret = spapr_populate_memory(spapr, fdt);
>> @@ -1613,6 +1614,10 @@ static int spapr_post_load(void *opaque, int
>> version_id)
>> }
>> }
>>
>> + if (spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
>> + spapr_xive_mmio_map(spapr);
>> + }
>> +
>> return err;
>> }
>>
>> diff --git a/include/hw/ppc/spapr_xive.h b/include/hw/ppc/spapr_xive.h
>> index f6d4bf26e06a..88355f7eb643 100644
>> --- a/include/hw/ppc/spapr_xive.h
>> +++ b/include/hw/ppc/spapr_xive.h
>> @@ -84,5 +84,6 @@ typedef struct sPAPRMachineState sPAPRMachineState;
>> void spapr_xive_hcall_init(sPAPRMachineState *spapr);
>> void spapr_xive_populate(sPAPRMachineState *spapr, int nr_servers, void
>> *fdt,
>> uint32_t phandle);
>> +void spapr_xive_mmio_map(sPAPRMachineState *spapr);
>>
>> #endif /* PPC_SPAPR_XIVE_H */
>
