Re: [Qemu-devel] [RFC v2 3/6] memory: support unmapping of MemoryRegion mapped into HVA parent

[Michael S. Tsirkin]

On Wed, Jun 17, 2015 at 05:04:37PM +0200, Igor Mammedov wrote:
> On Wed, 17 Jun 2015 10:14:00 +0200
> Paolo Bonzini <address@hidden> wrote:
> 
> > 
> > 
> > On 09/06/2015 12:08, Igor Mammedov wrote:
> > > 
> > > 3. until #2 is done we can't allow to map another memory region in
> > >    current flatview at the same range, hence need to keep list of
> > >    still active HVA ranges so we could check at 
> > > memory_region_add_subregion()
> > >    time that new mapping is allowed. => adding "Error **errp"
> > >    to memory_region_add_subregion() for reporting fail.
> > 
> > No, we cannot do that.  memory_region_add_subregion cannot fail.  The
> > RCU mechanism is transparent.
> So what are the options?
> 
> Actually memory_region_add_subregion(cannot fail) will continue to casue 
> problems
> in case one of the listeners fails and has no way to propagate error up the 
> stack.
> It could be that vhost_set_memory() or kvm_set_memslots() aren't able to 
> allocate
> memory for replacement table in low memory conditions and without
> way to report back QEMU is left with option to die.

mmap reserving memory just adds it to a data structure somewhere
within glibc, doesn't it? Looks like it's highly unlikely to fail,
so maybe killing QEMU if it does isn't a big deal.

> > 
> > Paolo