Re: mapping pages into new address spaces

Hmm, when you put it that way, I realize that my arguments probably  
aren't very compelling.  :)  But I'll say them anyway:

1. it makes more sense to me that if a pager knows it wants to do a  
mapping, it can just do it, instead of having to wait for a page  
fault.  This is an argument of "it will make the API and potentially  
some programs easier to understand."

2. I don't know the cost of handling page faults, but if they are at  
all expensive, proactively mapping pages will help avoid them.  This  
is an argument of performance.

I can understand if these arguments aren't compelling.

Thanks,
Josh

On Sep 9, 2006, at 12:27 PM, Bernhard Poess wrote:

> Whats the advantage of "your" approach to sending a startup ipc and  
> then having an IPC with a receive window pointing to the complete  
> address space and thus being able to map any page wherever you want  
> in the thread's address space ?
>
> -Bernhard
>
> On 9/8/06, Joshua Haberman <[email protected]> wrote:
> > Hi Josh,
> >
> >> Here's the only solution I can see: start the new thread  
> executing on
> >> a page that I *know* is unmapped, let it page fault on the first
> >> instruction, and then map the page on demand.  Is this the right  
> way
> >> to do it?  Am I thinking about this in the right way?
> >
> > Yes, you're right (except that I don't understand why you want a  
> page of
> > which you know that it's not mapped, in a newly created address  
> space,
> > this should be the case for every (user) page, I think).
> > Once the new thread's state is set to active, it waits for a  
> startup IPC
> > from its pager. This IPC contains IP and SP for the thread. Upon  
> receiving
> > the IPC, the thread will start running at the specified IP and  
> raise a
> > page fault. The pager can now reply with the needed page.
>
> Thanks for the replies -- I 'm glad to know I'm thinking about this  
> right.
> My next question would be: why couldn't the kernel automatically  
> accept
> any maps/grants that a thread receives from its pager, like it does  
> with
> maps/grants that are sent in reply to page faults?  It seems a bit
> roundabout that a pager can't proactively map pages into a thread's
> address space without cooperation from the thread in question.
>
> It seems logical that the maps sent in response to page faults should
> behave like unsolicited maps from a thread's pager.
>
> Josh
>
>
>