[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Status Bits and Super Pages
From: |
Marcus Brinkmann |
Subject: |
Re: Status Bits and Super Pages |
Date: |
Thu, 23 Aug 2007 13:54:56 +0200 |
User-agent: |
Wanderlust/2.14.0 (Africa) SEMI/1.14.6 (Maruoka) FLIM/1.14.8 (Shijō) APEL/10.6 Emacs/23.0.0 (i486-pc-linux-gnu) MULE/6.0 (HANACHIRUSATO) |
At Thu, 23 Aug 2007 13:20:00 +0200,
Neal H. Walfield wrote:
>
> Hi,
>
> I'm writing a memory manager and in order to make eviction decisions,
> I want to use the status bits. Because I'm a good citizen, I've
> written my memory manager to get the largest mappings possible from
> sigma0. My resource manager, however, sub-divides these pages and
> maps them to clients.
>
> I suspect (but please confirm) that my resource manager is only able
> to get the status bits for the mappings that it received from its
> pager, not the mappings it sent to its clients. That is, if it got a
> 64MB fpage from sigma0, even though it may have only handed out 4KB
> pages to clients, there is only a single status bit available to it:
> the one for the 64MB page. Is this interpretation correct?
As far as I remember it, yes, this is correct.
> If so, it seems that if I want status bits for the pages that the
> resource manager maps to its clients, then I have to be careful to
> only map pages of the same size as the ones the manager has. I
> imagine that I can request new mappings from sigma0 as appropriate,
Or insert an intermediate task.
> but this basically defeats the advantage of having mapped the
> superpages to begin with--most mappings will be small and thus the
> address space with quickly be filled with small mappings. Plus, with
> this approach, there is the overhead of the cost of the unmap and
> re-map operations.
Yes. However, there is a related problem that you can not unmap
selectively. I vaguely remember that Espen had a selective unmap
proposal at some time as a small modification to the L4 X.2 interface.
But I don't remember any details. Whatever that proposal was should
also provide an opportunity to do selective status bit checking,
shouldn't it?
Thanks,
Marcus