Re: [Qemu-devel] [PATCH v3 19/23] memory: per-AddressSpace dispatch

[Avi Kivity]

On 10/11/2012 10:45 AM, liu ping fan wrote:
> On Thu, Oct 11, 2012 at 4:31 PM, Avi Kivity <address@hidden> wrote:
>> On 10/11/2012 05:38 AM, liu ping fan wrote:
>>>>
>>>> +void address_space_init_dispatch(AddressSpace *as)
>>>> +{
>>>> +    AddressSpaceDispatch *d = g_new(AddressSpaceDispatch, 1);
>>>> +
>>>> +    d->phys_map  = (PhysPageEntry) { .ptr = PHYS_MAP_NODE_NIL, .is_leaf = 
>>>> 0 };
>>>> +    d->listener = (MemoryListener) {
>>>> +        .begin = mem_begin,
>>>> +        .region_add = mem_add,
>>>
>>> mem_add() has the fixed mapping relationship between addr(in
>>> system_memory) and mr. If we want to fold lookup logic of iommu into
>>> radix-tree, it will need external behavior. So I think
>>> address_space_init_dispatch(as) should be
>>> address_space_init_dispatch(as, listener).
>>
>> This is per address space, it's not tied to system_memory().
>>
> Then, where do we implement iommu translation logic? Above the radix-tree?

I will post patches soon.  Basically each iommu does the translation,
then uses another radix tree to continue the dispatch.  Eventually we
reach the bottom.

We can't use the radix tree to resolve iommu translations because they
change dynamically and are resolved on demand, while the radix tree is
read-mostly and prepared beforehand.

-- 
error compiling committee.c: too many arguments to function