Any reason we need memoryregionsection

Any reason we need memoryregionsection

[沈梦姣]

Usually when we add memory region in the platform init process, the corresponding address range in address space will not overlap with others, I’m not sure in which case we should use MemoryRegionSection? Thanks 

Re: Any reason we need memoryregionsection

[Peter Xu]

On Mon, Feb 22, 2021 at 10:55:15PM +0800, 沈梦姣 wrote:
> Usually when we add memory region in the platform init process, the corresponding address range in address space will not overlap with others, I’m not sure in which case we should use MemoryRegionSection? Thanks 
> 

Memory regions can overlap, see memory_region_add_subregion_overlap() and the
comment above its definition.  memory_region_add_subregion() is actually
similar to that, but just set the MR priority to zero as default.

IMHO MemoryRegion is the major abstraction layer of the address range in QEMU
along with the bigger AddressSpace idea, while MemoryRegionSection is mostly
used as a temporarly variable to describe part of a MemoryRegion in many
contexts, so that this partial range information can be further consumed by
callers of a lot of memory APIs.

Feel free to have a look at commit 2177955a89 ("memory: introduce
memory_region_find()", 2011-12-20) too - that should be the place where
MemoryRegionSection is firstly introduced.

Thanks,

-- 
Peter Xu

Re: Any reason we need memoryregionsection

[Peter Maydell]

On Mon, 22 Feb 2021 at 17:19, Peter Xu <peterx@redhat.com> wrote:
>
> On Mon, Feb 22, 2021 at 10:55:15PM +0800, 沈梦姣 wrote:
> > Usually when we add memory region in the platform init process, the corresponding address range in address space will not overlap with others, I’m not sure in which case we should use MemoryRegionSection? Thanks
> >
>
> Memory regions can overlap, see memory_region_add_subregion_overlap() and the
> comment above its definition.  memory_region_add_subregion() is actually
> similar to that, but just set the MR priority to zero as default.
>
> IMHO MemoryRegion is the major abstraction layer of the address range in QEMU
> along with the bigger AddressSpace idea, while MemoryRegionSection is mostly
> used as a temporarly variable to describe part of a MemoryRegion in many
> contexts, so that this partial range information can be further consumed by
> callers of a lot of memory APIs.

Yes. You only really need to care about MemoryRegionSections if you're
using an API that deals with them (notably if you're working with a
MemoryListener, or if you're using memory_region_find()). Most code,
and in particular almost all device and board code, works with
MemoryRegions.

-- PMM