Re: [Qemu-devel] [PATCH] hostmem: fix reference to uninit mr

[Peter Xu]

On Fri, Mar 10, 2017 at 10:17:30AM +0100, Peter Maydell wrote:
> On 10 March 2017 at 09:59, Peter Xu <address@hidden> wrote:
> > On Fri, Mar 10, 2017 at 09:33:57AM +0100, Paolo Bonzini wrote:
> >>
> >>
> >> On 10/03/2017 05:13, Peter Xu wrote:
> >> > Trying to get memory region size of an uninitialized memory region is
> >> > probably not a good idea. Let's just do the alloc no matter what.
> >> >
> >> > Signed-off-by: Peter Xu <address@hidden>
> >>
> >> What is the effect of the bug? The idea was to do the initialization
> >> once only (memory_region_size ought to be 0 when the MR is
> >> uninitialized; now it is ugly but it made more sense when MemoryRegion
> >> was just a C struct and not a QOM object).
> >
> > It's not really a bug. I just saw it, thought it was something not
> > quite right, so posted a patch.
> 
> We could reasonably abstract out the test into a function
> bool backend_mr_initialized(HostMemoryBackend *backend)
> {
>     /* We forbid zero-length in file_backend_memory_alloc,
>      * so zero always means "we haven't allocated the backend
>      * MR yet".
>      */
>     return memory_region_size(&backend->mr) != 0;
> }
> 
> and use it in file_backend_memory_alloc(), set_mem_path()
> and file_memory_backend_set_share(). That would make the intent
> clearer here I think.

Agree, maybe use it in hostmem.c as well where capable?

(Paolo: I wasn't planning to add anything there, but if any of you
 like me to do this cleanup, I would be glad to :)

-- peterx