On Tue, Mar 11, 2014 at 09:03:29AM +0100, Paolo Bonzini wrote:
> Il 11/03/2014 04:50, Hu Tao ha scritto:
> >On Tue, Mar 04, 2014 at 03:00:45PM +0100, Paolo Bonzini wrote:
> >>Like the previous patch did in exec.c, split memory_region_init_ram and
> >>memory_region_init_ram_from_file, and push mem_path one step further up.
> >>Other RAM regions than system memory will now be backed by regular RAM.
> >
> >This changes qemu's behaviour regarding using hugetlbfs, especially when
> >size of other RAM regions is significant compared to system memory. Will
> >this a problem?(compatibilities, user configurations...)
>
> I think it's a bugfix, especially with 1G hugepages. Wasting 1G on
> VGA memory is not really a good idea.
What about this
if (memory < hpagesize) {
return NULL;
}
> >>
> >>Also, boards that do not use memory_region_allocate_system_memory will
> >>not support -mem-path anymore. This can be changed before the patches
> >>are merged by migrating boards to use the function.
> >
> >IIUC, memory_region_allocate_system_memory() is only called once at
> >board initialization time. In most cases, it just works. But there
> >are cases that system memory is not initialized by a single call to
> >memory_region_allocate_system_memory() that total memory is splitted
> >into banks each of which is initialized individually, see
> >ppc4xx_sdram_adjust(). How to convert to
> >memory_region_allocate_system_memory() in this case? Should we map
> >banks into numa nodes?
>
> The solution would be to use aliases to split the system RAM into banks.
>
> Paolo
