Re: [Qemu-devel] [RFC 0/5] Memory transaction attributes API

[Peter Maydell]

On 18 March 2015 at 08:38, Edgar E. Iglesias <address@hidden> wrote:
> 2. The series introduces the read and write _with_attrs. Another
> approach could be to add a (for example) single .access callback.
> The callback could take a structure pointer as input, e.g:
>
> struct MemTx {
>     bool rw;
>     hwaddr addr;
>     int size;
>     uint64_t data;
>     MemTxAttrs attrs;
> }
>
> MemTxResult access(MemTx *tx)
>
> The benefit I see is that this will make it easier for us in the
> future to add new fields if needed.

I see the reasoning, but I looked at this and it just felt
awkward when I tried writing the code that way -- the code
calling the access function ends up having to create and
mutate this MemTx struct as it loops round calling the
access callback. In particular, memory.c has kept the
read and write paths separate all the way through, so
squashing them back together into a single callback
which will then immediately want to split them back out
into read vs write seemed a bit daft.

-- PMM