Re: [Qemu-devel] [Qemu-ppc] [PATCH qemu] ppc/spapr: Receive and store device tree blob from SLOF

[Alexey Kardashevskiy]

On 12/11/2018 20:05, Greg Kurz wrote:
> On Mon, 12 Nov 2018 15:12:26 +1100
> Alexey Kardashevskiy <address@hidden> wrote:
> 
>> On 12/11/2018 05:10, Greg Kurz wrote:
>>> Hi Alexey,
>>>
>>> Just a few remarks. See below.
>>>
>>> On Thu,  8 Nov 2018 12:44:06 +1100
>>> Alexey Kardashevskiy <address@hidden> wrote:
>>>   
>>>> SLOF receives a device tree and updates it with various properties
>>>> before switching to the guest kernel and QEMU is not aware of any changes
>>>> made by SLOF. Since there is no real RTAS (QEMU implements it), it makes
>>>> sense to pass the SLOF final device tree to QEMU to let it implement
>>>> RTAS related tasks better, such as PCI host bus adapter hotplug.
>>>>
>>>> Specifially, now QEMU can find out the actual XICS phandle (for PHB
>>>> hotplug) and the RTAS linux,rtas-entry/base properties (for firmware
>>>> assisted NMI - FWNMI).
>>>>
>>>> This stores the initial DT blob in the sPAPR machine and replaces it
>>>> in the KVMPPC_H_UPDATE_DT (new private hypercall) handler.
>>>>
>>>> This adds an @update_dt_enabled machine property to allow backward
>>>> migration.
>>>>
>>>> SLOF already has a hypercall since
>>>> https://github.com/aik/SLOF/commit/e6fc84652c9c0073f9183
>>>>
>>>> Signed-off-by: Alexey Kardashevskiy <address@hidden>
>>>> ---
>>>>  include/hw/ppc/spapr.h |  7 ++++++-
>>>>  hw/ppc/spapr.c         | 29 ++++++++++++++++++++++++++++-
>>>>  hw/ppc/spapr_hcall.c   | 32 ++++++++++++++++++++++++++++++++
>>>>  hw/ppc/trace-events    |  2 ++
>>>>  4 files changed, 68 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/include/hw/ppc/spapr.h b/include/hw/ppc/spapr.h
>>>> index ad4d7cfd97..f5dcaf44cb 100644
>>>> --- a/include/hw/ppc/spapr.h
>>>> +++ b/include/hw/ppc/spapr.h
>>>> @@ -100,6 +100,7 @@ struct sPAPRMachineClass {
>>>>  
>>>>      /*< public >*/
>>>>      bool dr_lmb_enabled;       /* enable dynamic-reconfig/hotplug of LMBs 
>>>> */
>>>> +    bool update_dt_enabled;    /* enable KVMPPC_H_UPDATE_DT */
>>>>      bool use_ohci_by_default;  /* use USB-OHCI instead of XHCI */
>>>>      bool pre_2_10_has_unused_icps;
>>>>      bool legacy_irq_allocation;
>>>> @@ -136,6 +137,9 @@ struct sPAPRMachineState {
>>>>      int vrma_adjust;
>>>>      ssize_t rtas_size;
>>>>      void *rtas_blob;
>>>> +    uint32_t fdt_size;
>>>> +    uint32_t fdt_initial_size;  
>>>
>>> I don't quite see the purpose of fdt_initial_size... it seems to be only
>>> used to print a trace.  
>>
>>
>> Ah, lost in rebase. The purpose was to test if the new device tree has
>> not grown too much.
>>
> 
> Ok, makes sense during development.
> 
>>
>>
>>>   
>>>> +    void *fdt_blob;
>>>>      long kernel_size;
>>>>      bool kernel_le;
>>>>      uint32_t initrd_base;
>>>> @@ -462,7 +466,8 @@ struct sPAPRMachineState {
>>>>  #define KVMPPC_H_LOGICAL_MEMOP  (KVMPPC_HCALL_BASE + 0x1)
>>>>  /* Client Architecture support */
>>>>  #define KVMPPC_H_CAS            (KVMPPC_HCALL_BASE + 0x2)
>>>> -#define KVMPPC_HCALL_MAX        KVMPPC_H_CAS
>>>> +#define KVMPPC_H_UPDATE_DT      (KVMPPC_HCALL_BASE + 0x3)
>>>> +#define KVMPPC_HCALL_MAX        KVMPPC_H_UPDATE_DT
>>>>  
>>>>  typedef struct sPAPRDeviceTreeUpdateHeader {
>>>>      uint32_t version_id;
>>>> diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
>>>> index c08130facb..5e2d4d211c 100644
>>>> --- a/hw/ppc/spapr.c
>>>> +++ b/hw/ppc/spapr.c
>>>> @@ -1633,7 +1633,10 @@ static void spapr_machine_reset(void)
>>>>      /* Load the fdt */
>>>>      qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
>>>>      cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
>>>> -    g_free(fdt);
>>>> +    g_free(spapr->fdt_blob);
>>>> +    spapr->fdt_size = fdt_totalsize(fdt);
>>>> +    spapr->fdt_initial_size = spapr->fdt_size;
>>>> +    spapr->fdt_blob = fdt;  
>>>
>>> Hmm... It looks weird to store state in a reset handler. I'd rather zeroe
>>> both fdt_blob and fdt_size here.  
>>
>>
>> The device tree is built from the reset handler and the idea is that we
>> want to always have some tree in the machine.
>>
> 
> Yes of course, I forgot that we need to keep the fdt to be kept
> somewhere so that we can use it :). My remark has more to do
> with migration actually: the fdt built at reset time is supposed
> to derive from the command line and hot-(un)plugged devices, ie,
> identical in source and destination. This isn't state we should
> migrate IIUC.

Having some device tree all the time seems more convenient than managing
the state when we do have one and when we do not.

It is not a big deal though, I'd wait and see what David thinks. Thanks,



> Maybe add a boolean field that tells that the fdt was updated, use
> it in spapr_dtb_needed() and reset it in spapr_machine_reset() ?
> 
>>
>>
>>>   
>>>>  
>>>>      /* Set up the entry state */
>>>>      spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
>>>> @@ -1887,6 +1890,27 @@ static const VMStateDescription 
>>>> vmstate_spapr_irq_map = {
>>>>      },
>>>>  };
>>>>  
>>>> +static bool spapr_dtb_needed(void *opaque)
>>>> +{
>>>> +    sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
>>>> +
>>>> +    return smc->update_dt_enabled;  
>>>
>>> This means we always migrate the fdt, even if migration occurs before
>>> SLOF could call KVMPPC_H_UPDATE_DT.
>>>
>>> With spapr->fdt_blob set to NULL on reset, a better check would be:
>>>
>>>     sPAPRMachineState *spapr = SPAPR_MACHINE(opaque);
>>>
>>>     return smc->update_dt_enabled && spapr->fdt_blob;
>>>   
>>>> +}
>>>> +
>>>> +static const VMStateDescription vmstate_spapr_dtb = {
>>>> +    .name = "spapr_dtb",
>>>> +    .version_id = 1,
>>>> +    .minimum_version_id = 1,
>>>> +    .needed = spapr_dtb_needed,
>>>> +    .fields = (VMStateField[]) {
>>>> +        VMSTATE_UINT32(fdt_initial_size, sPAPRMachineState),
>>>> +        VMSTATE_UINT32(fdt_size, sPAPRMachineState),
>>>> +        VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, sPAPRMachineState, 0, NULL,
>>>> +                                     fdt_size),
>>>> +        VMSTATE_END_OF_LIST()
>>>> +    },
>>>> +};
>>>> +
>>>>  static const VMStateDescription vmstate_spapr = {
>>>>      .name = "spapr",
>>>>      .version_id = 3,
>>>> @@ -1915,6 +1939,7 @@ static const VMStateDescription vmstate_spapr = {
>>>>          &vmstate_spapr_cap_sbbc,
>>>>          &vmstate_spapr_cap_ibs,
>>>>          &vmstate_spapr_irq_map,
>>>> +        &vmstate_spapr_dtb,  
>>>
>>> This needs to be rebased.
>>>
>>> <<<<<<<
>>>         &vmstate_spapr_cap_nested_kvm_hv,
>>> =======
>>>         &vmstate_spapr_dtb,  
>>>>>>>>>>  
>>>
>>>
>>> I'll try to find some time to respin the PHB hotplug series and I'll happily
>>> give a try to this patch.  
>>
>>
>> Good :)
>>
>>
>>>   
>>>>          NULL
>>>>      }
>>>>  };
>>>> @@ -3849,6 +3874,7 @@ static void spapr_machine_class_init(ObjectClass 
>>>> *oc, void *data)
>>>>      hc->unplug = spapr_machine_device_unplug;
>>>>  
>>>>      smc->dr_lmb_enabled = true;
>>>> +    smc->update_dt_enabled = true;
>>>>      mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
>>>>      mc->has_hotpluggable_cpus = true;
>>>>      smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
>>>> @@ -3965,6 +3991,7 @@ static void 
>>>> spapr_machine_3_0_class_options(MachineClass *mc)
>>>>  
>>>>      smc->legacy_irq_allocation = true;
>>>>      smc->irq = &spapr_irq_xics_legacy;
>>>> +    smc->update_dt_enabled = false;
>>>>  }
>>>>  
>>>>  DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
>>>> diff --git a/hw/ppc/spapr_hcall.c b/hw/ppc/spapr_hcall.c
>>>> index ae913d070f..d5833f3f8d 100644
>>>> --- a/hw/ppc/spapr_hcall.c
>>>> +++ b/hw/ppc/spapr_hcall.c
>>>> @@ -1717,6 +1717,36 @@ static target_ulong 
>>>> h_get_cpu_characteristics(PowerPCCPU *cpu,
>>>>  
>>>>      args[0] = characteristics;
>>>>      args[1] = behaviour;
>>>> +    return H_SUCCESS;
>>>> +}
>>>> +
>>>> +static target_ulong h_update_dt(PowerPCCPU *cpu, sPAPRMachineState *spapr,
>>>> +                                target_ulong opcode, target_ulong *args)
>>>> +{
>>>> +    target_ulong dt = ppc64_phys_to_real(args[0]);
>>>> +    struct fdt_header hdr = { 0 };
>>>> +    unsigned cb;
>>>> +    sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
>>>> +
>>>> +    cpu_physical_memory_read(dt, &hdr, sizeof(hdr));
>>>> +    cb = fdt32_to_cpu(hdr.totalsize);
>>>> +
>>>> +    if (fdt_check_full(spapr->fdt_blob, cb)) {
>>>> +        trace_spapr_update_dt_failed(spapr->fdt_initial_size, cb,
>>>> +            fdt32_to_cpu(hdr.magic));
>>>> +        return H_PARAMETER;
>>>> +    }
>>>> +
>>>> +    if (!smc->update_dt_enabled) {
>>>> +        return H_SUCCESS;
>>>> +    }
>>>> +
>>>> +    g_free(spapr->fdt_blob);
>>>> +    spapr->fdt_size = cb;
>>>> +    spapr->fdt_blob = g_malloc0(cb);
>>>> +    cpu_physical_memory_read(dt, spapr->fdt_blob, cb);
>>>> +
>>>> +    trace_spapr_update_dt(cb);
>>>>  
>>>>      return H_SUCCESS;
>>>>  }
>>>> @@ -1822,6 +1852,8 @@ static void hypercall_register_types(void)
>>>>  
>>>>      /* ibm,client-architecture-support support */
>>>>      spapr_register_hypercall(KVMPPC_H_CAS, h_client_architecture_support);
>>>> +
>>>> +    spapr_register_hypercall(KVMPPC_H_UPDATE_DT, h_update_dt);
>>>>  }
>>>>  
>>>>  type_init(hypercall_register_types)
>>>> diff --git a/hw/ppc/trace-events b/hw/ppc/trace-events
>>>> index dc5e65aee9..4432a5ce74 100644
>>>> --- a/hw/ppc/trace-events
>>>> +++ b/hw/ppc/trace-events
>>>> @@ -22,6 +22,8 @@ spapr_cas_pvr_try(uint32_t pvr) "0x%x"
>>>>  spapr_cas_pvr(uint32_t cur_pvr, bool explicit_match, uint32_t new_pvr) 
>>>> "current=0x%x, explicit_match=%u, new=0x%x"
>>>>  spapr_h_resize_hpt_prepare(uint64_t flags, uint64_t shift) 
>>>> "flags=0x%"PRIx64", shift=%"PRIu64
>>>>  spapr_h_resize_hpt_commit(uint64_t flags, uint64_t shift) 
>>>> "flags=0x%"PRIx64", shift=%"PRIu64
>>>> +spapr_update_dt(unsigned cb) "New blob %u bytes"
>>>> +spapr_update_dt_failed(unsigned cbold, unsigned cbnew, unsigned magic) 
>>>> "Old blob %u bytes, new blob %u bytes, magic 0x%x"
>>>>  
>>>>  # hw/ppc/spapr_iommu.c
>>>>  spapr_iommu_put(uint64_t liobn, uint64_t ioba, uint64_t tce, uint64_t 
>>>> ret) "liobn=0x%"PRIx64" ioba=0x%"PRIx64" tce=0x%"PRIx64" ret=%"PRId64  
>>>   
>>
> 

-- 
Alexey