Re: memory_object_lock_request and memory_object_data_return fnord

[Neal H Walfield]

> > In the scenario case, we do a completely unnecessary read!  That is
> > not demand paging; that is a waste.
> 
> Oh, you're talking about the case where we are writing complete pages?
> Yes, that's a defect of any scheme like this.  In principle, using the
> vm_copy interface is the only thing that can effectively avoid it.

I think that that optimization could be easily integrated into
pager_memcpy.

> > But the manager knows; why force the kernel to guess?  Say we send:
> > 
> >         io_read (file, data, vm_page_size * 4, 0, &amount)
> > 
> > By the time the kernel detects a sequential read, it is already too
> > late to be of any use.
> 
> Well, this is an old Mach debate.  :)
> 
> But note that the *manager* doesn't know any better than the kernel.
> Only the *user* knows what the access pattern is.  (Even if the "user"
> resides in the same task as the pager.)

Ah right.

> So we might someday want a way for users to declare things about
> memory regions mapped in their address space, some kind of declaration
> like that might be useful.  That would be the time to add msync too.

But, since the user, in this case, is in the same task, we can do some
black magic.  Do you think that that is reasonable?