On Tue, 12 Apr 2016 23:45:18 +0200
Paolo Bonzini <pbonz...@redhat.com> wrote:
>
>
> On 07/04/2016 23:31, Dominik Dingel wrote:
> > diff --git a/exec.c b/exec.c
> > index c4f9036..1ae98e4 100644
> > --- a/exec.c
> > +++ b/exec.c
> > @@ -1241,6 +1241,7 @@ static void *file_ram_alloc(RAMBlock *block,
> > void *area;
> > int fd = -1;
> > int64_t page_size;
> > + int64_t alignment;
> >
> > if (kvm_enabled() && !kvm_has_sync_mmu()) {
> > error_setg(errp,
> > @@ -1296,7 +1297,8 @@ static void *file_ram_alloc(RAMBlock *block,
> > }
> >
> > page_size = qemu_fd_getpagesize(fd);
> > - block->mr->align = page_size;
> > + alignment = MAX(page_size, QEMU_VMALLOC_ALIGN);
> > + block->mr->align = alignment;
> >
> > if (memory < page_size) {
> > error_setg(errp, "memory size 0x" RAM_ADDR_FMT " must be equal to "
> > @@ -1305,7 +1307,7 @@ static void *file_ram_alloc(RAMBlock *block,
> > goto error;
> > }
> >
> > - memory = ROUND_UP(memory, page_size);
> > + memory = ROUND_UP(memory, alignment);
>
> I think this change is not necessary either; it is enough to change the
> qemu_ram_mmap below.
You are right "memory" will never be back propagated to anything related to
memory slots.
On which we have such a size constraint.
Thanks,
Dominik
