Re: [Qemu-devel] [PATCH v3 06/22] memory-device: document MemoryDeviceClass

[Igor Mammedov]

On Thu, 20 Sep 2018 12:32:27 +0200
David Hildenbrand <address@hidden> wrote:

> Document the functions and when to not expect errors.
> 
> Signed-off-by: David Hildenbrand <address@hidden>
> ---
>  include/hw/mem/memory-device.h | 13 +++++++++++++
>  1 file changed, 13 insertions(+)
> 
> diff --git a/include/hw/mem/memory-device.h b/include/hw/mem/memory-device.h
> index f02b229837..d6853156ff 100644
> --- a/include/hw/mem/memory-device.h
> +++ b/include/hw/mem/memory-device.h
> @@ -29,9 +29,22 @@ typedef struct MemoryDeviceState {
>      Object parent_obj;
>  } MemoryDeviceState;
>  
> +/**
> + * MemoryDeviceClass:
> + * @get_addr: The address of the @md in guest physical memory. "0" means that
> + * no address has been specified by the user and that no address has been
> + * assigned yet.
While looking at attempts to implement
 '[Qemu-devel] [PATCH v12 6/6] tpm: add ACPI memory clear interface'
on QEMU side, where we are trying to clear RAM going via ramblocks list.

Maybe we are doing it backwards and instead of trying to deal with host
abstraction RAMBlocks/MemoryRegion and follow up efforts to tag it with
device model attributes
  '[PATCH] RFC: mark non-volatile memory region'
we should do it other way around and approach it from guest point of view
like firmware would do, i.e.
make TPM enumerate RAM devices and tell them to clear related memory.
Concrete device would know how to do it properly and/or how to ask backend
to do it without need to tag RAMBlocks with device model specific info.
That would allow to enumerate all RAM without trying to filter out not RAM
RAMBlocks and not pollute host specific RAMBlock/MemoryRegion with device
specific flags.

However we can't do it right now, since built-in memory doesn't have
a corresponding device model and we use MemoryRegions directly.
Maybe we should use memory-devices and create derivative 'ramchip' device
to represent builtin RAM and use it instead.

So if we were to do that, we would need to make get_addr() return value 0
a valid one (suggest (uint64_t)-1 as unset value) and/or make built in
memory work outside of device_memory region as short cut impl.