Re: [Qemu-devel] [PATCH v4 1/2] arch_init: align MR size to target page size

[Peter Maydell]

On 19 August 2013 18:37, Laszlo Ersek <address@hidden> wrote:
> ((3) memory_region_size() is slightly different from
> int128_get64(mr->size); it has a special case for int128_2_64() -- and I
> don't understand that.

The special case is because valid memory region sizes range
from 0 to 2^64, *inclusive*. [2^64-sized regions tend to be
containers representing address spaces like PCI or the system
memory space.] Inside memory.c we store the size
in an int128 in the way you'd expect. However some of the
memory region APIs take or return the size of the region
as a uint64_t, with the convention that UINT64_MAX means
"actually 2^64" (with a size of really 2^64 - 1 not being
valid). So the special case here is doing the conversion
from an int128 representation of the size to the uint64_t
representation. You can see the same thing going the other
way in memory_region_init(), where we take the size as a
uint64_t and convert it to an int128 with

    mr->size = int128_make64(size);
    if (size == UINT64_MAX) {
        mr->size = int128_2_64();
    }

(Note that int128_get64() of an int128 which == 2^64 or
more will assert, so int128_get64(mr->size) is a bit of a
code smell; it happens to be OK here because we know that a
RAM-backed MR will never be a 2^64-sized region.)

-- PMM