Re: [PATCH v2 07/22] intel_iommu: add set/unset_iommu_context callback

[Auger Eric]

Hi Yi,

On 3/31/20 2:25 PM, Liu, Yi L wrote:
> Hi Eric,
> 
>> From: Auger Eric < address@hidden>
>> Sent: Tuesday, March 31, 2020 4:24 AM
>> To: Liu, Yi L <address@hidden>; address@hidden;
>> Subject: Re: [PATCH v2 07/22] intel_iommu: add set/unset_iommu_context 
>> callback
>>
>> Yi,
>>
>> On 3/30/20 6:24 AM, Liu Yi L wrote:
>>> This patch adds set/unset_iommu_context() impelementation in Intel
>> This patch implements the set/unset_iommu_context() ops for Intel vIOMMU.
>>> vIOMMU. For Intel platform, pass-through modules (e.g. VFIO) could
>>> set HostIOMMUContext to Intel vIOMMU emulator.
>>>
>>> Cc: Kevin Tian <address@hidden>
>>> Cc: Jacob Pan <address@hidden>
>>> Cc: Peter Xu <address@hidden>
>>> Cc: Yi Sun <address@hidden>
>>> Cc: Paolo Bonzini <address@hidden>
>>> Cc: Richard Henderson <address@hidden>
>>> Cc: Eduardo Habkost <address@hidden>
>>> Signed-off-by: Liu Yi L <address@hidden>
>>> ---
>>>  hw/i386/intel_iommu.c         | 71
>> ++++++++++++++++++++++++++++++++++++++++---
>>>  include/hw/i386/intel_iommu.h | 21 ++++++++++---
>>>  2 files changed, 83 insertions(+), 9 deletions(-)
>>>
>>> diff --git a/hw/i386/intel_iommu.c b/hw/i386/intel_iommu.c
>>> index 4b22910..fd349c6 100644
>>> --- a/hw/i386/intel_iommu.c
>>> +++ b/hw/i386/intel_iommu.c
>>> @@ -3354,23 +3354,33 @@ 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));
>>>          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 +3446,55 @@ 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 *iommu_ctx)
>>> +{
>>> +    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];
>>> +
>>> +    assert(!vtd_dev_icx);
>>> +
>>> +    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->iommu_ctx = iommu_ctx;
>>> +
>>> +    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);
>>> +    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 +3790,8 @@ 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..b5fefb9 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 *iommu_ctx;
>> I don't get why we don't have standard QOM inheritance instead of this
>> handle?
>> VTDHostContext parent_obj;
>>
>> like IOMMUMemoryRegion <- MemoryRegion <- Object
> 
> Here it is not inherit the object. It's just cache the HostIOMMUContext
> pointer in vIOMMU. Just like AddressSpace, it has a MemoryRegion pointer.
> Here is the same, VTDHostIOMMUContext is just a wrapper to better manage
> it in vVT-d. It's not inheriting.

Yep I've got it now ;-)
> 
>>> +    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];
>> At this point of the review, it is unclear to me why the context is
>> associated to a device.
> 
> HostIOMMUContext can be per-device or not. It depends on how vIOMMU
> manage it. For vVT-d, it's per device as the container is per-device.
> 
>> Up to now you have not explained it should. If
>> so why isn't it part of VTDAddressSpace?
> 
> Ah, I did have considered it. But I chose to use a separate one as
> context is not really tied with an addresspace. It's better to mange
> it with a separate structure.

OK

Thanks

Eric
> 
> Regards,
> Yi Liu
>